Disponible en ligne sur

ScienceDirect
www.sciencedirect.com
IRBM 38 (2017) 305–324

General Review

Stark Assessment of Lifestyle Based Human Disorders Using Data Mining

Based Learning Techniques
M. Sharma a,∗ , , G. Singh b , R. Singh b
a Dept. of Computer Sc. and App., DAV University, Jalandhar, India
b DCS, GNDU, Amritsar, Punjab, India

Received 5 April 2017; received in revised form 20 August 2017; accepted 15 September 2017
Available online 18 October 2017

Highlights Graphical abstract

• Lifestyle based human disor-

ders diagnosed have been done
by using different data mining
techniques.
• Common and live datasets for
lifestyle based human disorders
have been mentioned.
• Rate of accuracies achieved us-
ing different mining techniques
is also highlighted.
• Effect of preprocessing, re-
lationship between disease,
datatype and mining approach
is analyzed.
• A novel hybrid diagnosis model
for lifestyle based human disor-
ders is proposed.

Abstract
Background: Medical informatics has observed an unrestrained growth in the database. Latest advancements in the field of medical sciences
have wiped out lots of critical diseases. Nowadays, the medical industry is affluent in data sources. These data sources are of use only if these are
effectively analyzed on time.
Methods: Data mining techniques are artificially intelligent and used to investigate known and unknown patterns available in the medical
databases. Nowadays, data mining techniques are chronically used to mine abundant data sources of medical science. This paper explores the
practice of diverse data mining techniques, the role of dataset used, effect of preprocessing, and the performances of different data mining
techniques in diagnosis of different lifestyle based diseases. The venture of this paper is to fetch out stark assessments of different data mining
techniques used in medical sciences.
Results: By far, surveillance discloses that significant effort has been made for mining the data allied to the Cardiology and Diabetes. As per
Google Scholar, in last seven years, the percentage of articles published related to cardio, diabetes, digestive, dentistry and ophthalmology disease

* Corresponding author.
E-mail address: manik10143@davuniversity.org (M. Sharma).

https://doi.org/10.1016/j.irbm.2017.09.002
1959-0318/© 2017 AGBM. Published by Elsevier Masson SAS. All rights reserved.
306 M. Sharma et al. / IRBM 38 (2017) 305–324

diagnosis using data mining are 42%, 26%, 18%, 10% and 4% respectively. So, a little attention has been paid to develop predictive model for the
diseases viz. ophthalmology, dentistry and digestive disorders. In addition, the rate of usage of preprocessing in diagnosis of different disorders
related to cardio, diabetes, digestive, dentistry and ophthalmology lies between 10.65%–17.75%, 8.48%–14.80%, 4.58–8.93%, 2.96%–7.73% and
5.83%–12.93% respectively.
Conclusion: An attention is obligatory to develop smart diagnostic system to aware and save human masses from wide critical spectrum of
diseases related to ophthalmology, oral and digestive systems.
© 2017 AGBM. Published by Elsevier Masson SAS. All rights reserved.

Keywords: Data mining; Lifestyle; Heart disease; Diabetes; Ophthalmology; Oral and digestive disorder

1. Introduction to data mining
In modern times, Data mining has acknowledged an im-
mense deal of attention. Data mining is an automatic extraction
of patterns or useful information. It is used to stumble on the
consequence of the problem by probing the available facts and
data. It helps in measuring the association, nature and degree of
correlation in the different parameters of the dataset. Moreover,
the dataset can also be classified into several classes. At core
level, data mining is known as knowledge discovery database.
One can also predict the future by carefully analyzing the past
and current state of dataset.The commonly used data mining
techniques are naïve bayes, neural network (NN), genetic al-
gorithm (GA), support vector machine (SVM), decision tree
induction (DTI) etc. The complete working of data mining tech-
niques is based upon two sets of data i.e. Training and Testing.
Data mining is a difficult process as one has to train the system
regarding the characteristics or the features that have to be ex-
tracted. Data mining is one of the dominant research fields and
is used in almost all streams viz. Agriculture, Computer Sci-
ence, Mathematics, Chemistry, Finance, Economics, Medical
Science, Zoology, Bio-informatics etc. The major applications
of data mining are given below [1,2]:
• Agriculture: To study characteristics of soil, crop yield
analysis, optimizing use of pesticide, diagnosis of plant dis-
eases.
• Medical Science: Early diagnosis and prediction of various
diseases.
• Computer Science: Image processing, Network Security,
Computer Security.
• Business: Airline, Shopping Malls, Banks, Marketing.
• Natural and Life Sciences: Creation of new hypothesis,
Stimulation of new facts, Analyzing Protein structure etc.
• Engineering; CAD, CAM, Fault detection in production
lines.
Data mining [3] is a multistep approach. In the first phase,
data is collected from heterogeneous sources and is converted
into a homogeneous format. Different preprocessing and nor-
malization procedures are used to reduce data inconsistency. In
second phase, the data mining procedures are applied to exca-
vate some meaningful information. Third phase analyzes the
processed data and represent it in standardized format. Finally,
the upshots of data mining progression are used in decision
making process. In general, data mining is used to test hypoth-
esis or to discover some new or hidden patterns. Traditionally,
it was associated with hypothesis testing only. The idea was to
first prepare a statement that has to be tested against particular
set of data and condition.
There are certain anomalies [4] in mining the database. The
most imperativecrisis in data mining is the lack of long term vi-
sion. Sometimes, data used in mining is incorrect or may not
be updated. Moreover, some of the organization or departments
within an organization do not want to reveal the nature and con-
tents of their data. In addition, legal and privacy condition may
further create a problem in data mining. One of the crucial fac-
tors of decision making is having the right information at the
right time. Today, data collection is not of major issue. Rather,
analysis of data is of concern. Nowadays, the survival of organi-
zation is based upon the effectiveness in generating information
from their data. It is believed that the future and success of med-
ical industry is heavily dependent upon the data mining process.
This paper briefly summarized the endeavor of different re-
searchers for mining information for medical industry.
2. Related work
Suvarna Pawar and Smita Sikchi [5] have carried out a sur-
vey related to diabetes. The survey is focused on to unearth out
intricacy with the existing diagnosis systems. Authors found
that still it seems too difficult to diagnose diabetes with high
precision. Authors tried to emphasize on diverse classification
techniques that were used in diabetes diagnosis. Authors ob-
served that for PIMA data set, SVM seems to be a good solution
for early diagnosis of diabetes with 97% of accuracy.
Shubpreet Kaur, R.K. Bawa [6] stated that the use of data
mining provides an efficient way to mine the required clinical
information from capacious, raw and heterogeneous data. One
may use the techniques viz. decision tree, naïve bayes, logis-
tic regression and ANN to forecast different medical problems
with low cost and high precision. In addition, author presented
how data mining can be used to ascertain the interactions be-
tween health status and a disease.
Divya Tomar, Sonali Aggarwal [7] discovered the efficacy of
an assortment of data mining techniques like association, clas-
sification, clustering and regression in health sphere. Authors
briefly explained different data mining techniques with their
intrinsic worth and demerits. Authors suggested that before ap-
plying the classification techniques, one should preprocess the
 M. Sharma et al. / IRBM 38 (2017) 305–324
data i.e. data should be normalized by removing all types of
redundancies as it may degrade the execution time. Authors
recommended cross validation method in classification process.
Clustering is beneficial if there is no or missing information.
Finally, authors suggested to use a fusion of classification, clus-
tering and association to get better mining performance.
Punam Bajaj, Preeti Gupta [8] have studied the use of dis-
tinct data mining techniques in the diagnosis of heart disease.
Authors found the significant effort has been conceded out by
different researchers to diagnose heart related disease by us-
ing different classification and clustering techniques, feature
selection methods, fuzzy logic, naïve bayes, association rule,
regression techniques and their amalgamation. Authors have
also proposed a hybrid (GA and ANN) diagnosis system for
the same.
Parvez Ahmad et al. [9] have analyzed the convenience of
data mining techniques in healthcare. Authors mentioned that
data mining techniques plays a noteworthy role in nurturing
the momentous volume of data into useful information. Au-
thors discussed the role of classification, clustering, SVM, NN
and bayesian methods in mining the data related to breast,
lung cancer, heart diseases, smoking behavior, thyroid, dengue,
alzheimer’s, diabetes etc. Authors also mentioned some of the
challenges that researcher have faced while mining the data
related to healthcare industry. Author suggested that medical
industry has to focus on developing better and accurate med-
ical information system by using different data mining tech-
niques.
3. Research objective
The major objective of this research work is to show the
contribution of data mining in assessing the lifestyle based dis-
eases. The effort is made to review the existing literature of
different key authors whose research work is contributed for
both patients as well as the experts. The key points (disease,
methodology, results, accuracy) of the different research works
along with the use of tool or techniques are highlighted. Fi-
nally, focus is to determine the areas that require more attention
of data mining techniques. The existing literature is studied to
respond the following queries:
• What is longevity or civilization disease?
• Uncover association between eating habits, social and eco-
nomic behavior with lifestyle based diseases.
• To reveal the significance of data mining in early diagnosis
of civilization diseases.
• Contribution of data mining in analyzing data related to
heart patients.
• Significance of research work in scrutinizing data related to
diabetic patients.
• Highlighting some important healthcare data sets.
• Analyzing the need and effect of data preprocessing in
healthcare diagnostic systems.
• Identifying diseases where a significant data mining has
been carried out.
307
• Highlighting the domain of medical science where data
mining is still required.
• Need of predictive model for correcting lifestyle based dis-
orders in ophthalmology, oral and digestive system.
4. Lifestyle based diseases
Every individual has its own way of living that he/she patent
in coping with their physical, psychosomatic, societal, and
monetary environments on a regular basis. Lifestyle reflects a
person’s attitude, behavior, eating habits, social and economic
values. It is a reflection of the person that they perceive and
want to be seen by society. The lifestyle of a person is very
much affected by the genes, culture, society and the region. In
general, there are two major pillars of lifestyle viz. Eating habits
(intake of dietary, sugary products, alcohol consumption, smok-
ing etc.) and Social/Economic Behavior (family background,
social environment, economic conditions, employment, work-
ing conditions etc.). A vigorous or morbid lifestyle will most
likely be diffused across generations. Case et al. [10] observed
that there are around 27% chances that a child will adopt the
same lifestyle of his/her parents.
The bad lifestyle (Sedentary, laziness, alcohol, tobacco,
smoking, drug, overdose of dairy products, oil, sugary prod-
ucts) of a person may leads to several human disorders (Cardio-
vascular, diabetes, oral, ophthalmology, digestive) also known
as lifestyle diseases. In adults, generally, the social and eco-
nomic behavior are responsible for lifestyle based diseases. The
horrific lifestyle of an individual have significant role in oral
diseases. Some of the major longevity diseases are given be-
low:
– Cardiovascular
– Diabetes
– Oral
◦ Cancer
◦ Gum disease
◦ Tingling
◦ Trauma
◦ Bad breath
◦ Temporomandibular disorder
– Ophthalmology
◦ Glaucoma
◦ Diabetic retinopathy
◦ Cataract
– Digestive System
◦ Stomach cancer
– Neurology
◦ Stroke
◦ Autism
This paper briefly summarized the endeavor of different re-
searchers for mining information for lifestyle based human dis-
orders.
308 M. Sharma et al. / IRBM 38 (2017) 305–324
Fig. 1. Working principle of data mining classification process.
5. Data mining techniques (supervised learning)
The concept of data mining has been originated from three
different techniques viz. Statistics, Artificial Intelligence and
Machine Learning. Several heuristics have been projected to
perk up the competence of data mining process. Classification
is a dominant data mining technique. In general, classification
is categorized as single or multi class. In single class, there is
only one class label that has to be recognized. The elements that
belong to the class are known as normal and rest of the elements
are categorized as anomalies. The working principle of classi-
fication procedures is based upon training and testing data sets.
The working of classification model is represented in Fig. 1.
In the first phase, system is trained by using existing labeled
data. The training data set is then used to segregate the data
into diverse categories based upon the parameters and results of
the training data set. In other words, training data is one of the
major phases of classification process and is intended to impart
some sort of intelligence based upon which the data is mined.
Technically, training is a machine learning process. The optimal
mining of data is based upon the nature and level of trained
data set. Training data set is used to develop a classification or
predictive model.
To shun over fitting predicament and to distill the classifi-
cation model, sometimes, training data set is decomposed into
training set and validation set. The responsibility of validation
set is to perk up the performance of the under construction
framework. Finally, the accuracy of the developed system is
tested with test data. The accuracy can be measured by using
confusion matrix which represents the accuracy of recognizing
data set as true negative, true positive, false positive and false
negative. There are several classification techniques. The com-
monly used classification approaches are given below.
Decision tree induction [3,11] assists in erudition the con-
cepts and working of decision trees. Decision tree is a hierar-
chical constitution consisting of leaf, non-leaf nodes and edges.
The testing condition is represented by non leaf node. Edge
is used to represent the result. Decision tree classification ap-
proach is painless to implement. However, for a complex classi-
fication problem it requires lots of training data. The amount of
data is increased as the dimension of classification is increased.
Class labels are recognized by leaf nodes. ID3, C4.5 and CART
are some of milestones of DTI. Some of the major characteris-
tics of DTI are
• No need of domain knowledge.
• Suitable for investigative knowledge
Naive bayes [12] classification scrounges its idea from statis-
tics and probability. It is based upon posterior probability and
prior probability. One of the foremost anomalies in this ap-
proach is when one gets zero probability. To handle zero proba-
bility case, the concept of ‘Laplace’ estimation was introduced.
The best part of this approach is that in general it has minimum
rate or error.
Rule based classification [13] approach is based upon set of
rules. The rules can be represented in the form of ‘IF-Then’,
extracted from decision tree or can be generated by using a
sequential covering approach. The working principle of rule
based technique is based upon antecedent (Left side of rule)
and consequent (right side of rule). One of the major flaws of
rule based approach is sharp cut off. To overcome sharp cut off
problem, the concept of fuzzy logic is used.
SVM (Support Vector Machine) [14] is used to classify both
linear and non-linear types of data. In SVM, the dimensions of
training data are expanded by using non linear mapping. Credit
of SVM goes to Vladimir Vapnik. One of the best aspects of
SVM is that it leads to the high rate of accuracy. It is effectively
applicable for both prediction and classification process. It is
prominently used in digit recognition, voice identification and
object recognition. SVM is strongly based upon hyper plane.
It is used to differentiate elements from different classes. For
linear data, maximal marginal hyper plane is represented as

k
d(X ) =
T
Ci αi Si X T + b0
i=1
Here, Ci is class label. αi is Lagrangian multiplier. Si is support
vector. X T is test data.
Neural network [15] is a soft computing technique that
scrounges its idea from human mind. These are used for both
single and multi class problems. It was appeared in 1943. How-
ever, came into action in 1980s. Besides, solving the problem,
neural network can also learn from the former system or appli-
cations. NN are self organizing and adaptive in learning.
‘Genetic Algorithm’ [16,17] borrows its essential features
from natural genetics and let a population serene of numer-
ous individual chromosomes to grow under demarcate selec-
tion rules to breed a state which is optimizing the objective
function. It generally employs some heuristics like ‘Selection’,
‘Crossover’, and ‘Mutation’ to develop better solutions. GA is
capable of being applied to an enormously wide range of prob-
lems like Task Scheduling, Image Processing, Machine Learn-
ing, Data Mining, Medical Sciences etc. It starts its working
from a set of solutions rather than a single solution. The initial
population is generated randomly. Each solution of the problem
is adequately represented by encoding a chromosome. Every
 M. Sharma et al. / IRBM 38 (2017) 305–324
Fig. 2. Attributes of datasets for cardio, diabetes and cancer diagnosis.
chromosome has a fitness value allied with it. The anthology
of chromosomes is called population. The population at a par-
ticular instance is called generation. The genetic properties of
two chromosomes are blended to breedhealthier offspring us-
ing an operation called crossover. Mutation operation is used
to modify the behavior of the generated child to make it more
effective.
The performance of data mining process is measured by
finding the confusion matrix. There are four major parameters
of confusion matrix that indicates the efficiency of the tech-
nique in finding true positive, true negative, false positive and
false negative cases.
6. Data mining techniques and their role in medical science
No doubt, medical informatics is changing the scenario of
medical industry. With this advancement, one is able to diag-
nose and cure the problem effectively. The latest vaccination
and other treatment methodology has exterminated lots of fatal
diseases. However, there are still number of chronic diseases
(Cardiovascular, Respiratory, Malignant, Digestive, Diarrheal,
Depression, Malaria, Cancer, Diabetes etc.) that don’t appear to
affirm vanish. Number of researcher has studied the use of data
mining techniques and their role in medical science. The re-
maining part of section give brief detail about the data mining
techniques and upshots of the research carried out in medical
field. Nowadays, heart disease, diabetes, brain stroke, cancer
are effecting people badly.
6.1. Datasets used in disease diagnosis
A disease diagnostic system in healthcare application must
be trained through existing healthcare data. The data for train-
ing diagnostic system can be generated through different modes
like screening, physical and clinical diagnosis [18]. This data
can be represented in the form of text, numeric, sound, images
and signals. For disease diagnostic system, initially, the data is
309
required to train the system. The training data can be collected
from hospitals, clinics, research centers and online repositories.
It is difficult to collect data from hospitals and clinics. However,
one can easily obtain required data set from online reposito-
ries. UCI (UC Irvine) is an imperative online repository that
contains variety of data sets related to different domains. Some
of the important multivariate and Time-series data related to
lifestyle based human disorders are available on UCI reposi-
tory. Number of researcher have used these data sets for their
experimentation and analysis [19] (Table 1).
From existing research, it is observed that generally, dia-
betes patient’s data are represented in form of text and numeric
attributes. However, some of the researchers have also used
tongue images in diabetes diagnosis. Moreover, signal and im-
ages are most common media for representing the details of
cardio and cancer patients. Some of the important and common
parameters which are generally used in early diagnosis of dif-
ferent lifestyle based human disorders are mentioned in Fig. 2.
Table 2 represents the brief details of attributes, instances,
type and sources of data used in early diagnosis of lifestyle
based human disorder.
6.2. Diabetes diagnosis using classification approaches
Diabetes is one of a silent killer disease. Some of contribut-
ing factors behind diabetes are lack of physical activities, seden-
tary problems and obesity. Billions of people across the world
are affected by this so called modern society disease. An im-
proper treatment of diabetes may further leads to lots of other
physical problems and even death in some cases. In general,
diabetes are divided into three categories viz. Type 1, Type 2
and Gestational diabetes. Number of researchers has tried to
develop prediction model to deal with this life threatening dis-
ease. The remaining part of this section briefly describes the
contribution of different researchers in this domain.
Iyer, Jeylatha et al. [21] have performed a study to diag-
nose the diabetic using different classification techniques. Au-
thors stated that diabetes is one of the dominating diseases that
310 M. Sharma et al. / IRBM 38 (2017) 305–324

Table 1
Common datasets for lifestyle based human disorder diagnosis.
Data set Characteristic Number of instances Attribute characteristics Number of attributes
PIMA Multivariate 768 Integer, Real 8
130-US hospital Multivariate 100000 Integer 55
AIM ’94 Multivariate, Time-Series Categorical, Integer 20
Cardiotocography Multivariate 2126 Real 23
Heart Disease Multivariate 303 Categorical, Integer, Real 75
Echocardiogram Multivariate 132 Categorical, Integer, Real 12
PAMAP2 Multivariate, Time-Series 3850505 Real 52
SPECT Multivarite 267 Categorical 22
SPECTF Multivariate 267 Integer 44
Statlog Multivariate 270 Categorical, Real 13
Breast Cancer Multivariate 286 Categorical 9
Breast Cancer Wisconsin Multivariate 569 Real 32
Breast Tissue Multivariate 106 Real 10
Lung Cancer Multivariate 32 Integer 56
Liver Disorders Multivariate 345 Categorical, Integer, Real 7
Indian Liver Patient Dataset Multivariate 583 Integer, Real 10

Table 2
Datasets and their attributes used in earlier disease diagnosis research.
Authors Instances Attributes Type of data Source of data
Ilayaraja M., Meyyappan [20] 1000 19 Text and numeric Not mentioned
Aiswarya Iyer, S. Jeyalatha and Ronak 768 8 Integer, Real PIMA, UCI
Sumbaly [21]
Haofan Yang, Yi-Ping Phoebe Chen 500 06 Lung cancer cases, pathology Cancer Genome Atlas
[22] and clinical report
B. Venkatalakshmi, M.V. Shivsankar 294 13 Text and numeric UCI
[23]
Abhishek Taneja [24] 7339 15 and 8 Text and numeric PGI Chandigarh
K. Rajesh and V. Sangeetha [25] 760 9 Integer, Real PIMA, UCI
Soni and Ansari [26] 909 14 Text and numeric Cleveland Heart Disease database
R. Ramani et al. [27] 322 13 Images MIAS
R. Delshi Howsalya Devi, M. Indra 303 14 Text and numeric Cleveland (UCI)
Devi [28]
Jianfeng Zhang et al. [29] 827 23 Tongue images TCM hospital, Shanghai
Rajesh Kumar, Rajeev Srivastava, and 1000 115 Microscopic biopsy images Histology Image Dataset
Subodh Srivastava [30]
K. Vimala and Dr. V. Kalaivani [31] 171 Not mentioned ECG Bio signal Hospital
Prakash D., Uma Mageshwari T., Not mentioned Not mentioned Phonocardiogram Signals Not mentioned
Prabakaran K., and Suguna A. [32]
Hang Wu, Sahong Kim, and Keunsung 325 10 Heart Sound Signals Clinical audio CD
Bae [33]
Vishakha Pareek, R.K. Sharma [34] 160 Not mentioned Voice signal Not mentioned
Du-Yih Tsai and Yongbum Lee [35] 90 Not mentioned Ultrasonic images Gifu University Hospital
P. Rajeswari, G. Sophia Reena [36] 345 7 Textual and numeric UCI
Gouda I. Salama, M.B. Abdelhalim, 699 11 Image Wisconsin Breast Cancer
Magdy Abd-elghany Zeid [37] (Original), UCI
Gouda I. Salama, M.B. Abdelhalim, 569 32 Image Wisconsin Diagnosis Breast
Magdy Abd-elghany Zeid [37] Cancer
Gouda I. Salama, M.B. Abdelhalim, 198 34 Image Wisconsin Prognosis Breast Cancer
Magdy Abd-elghany Zeid [37]
Disha Sharma, Gagandeep Jindal [38] 1000 Not mentioned Images Lung Image Database Consortium
A.S. Aneesh Kumar, C. Jothi 2453 15 Text and numeric Sir Ivan Stedeford Hospital,
Venkateswaran [39] Chennai
Santi Wulan Purnami [40] 768 13 Text and numeric UCI (PIMA)
Santi Wulan Purnami [40] 2701 13 Text and numeric V.A. Medical Center, Cleveland
Clinic Foundation
 M. Sharma et al. / IRBM 38 (2017) 305–324
have affected millions of people worldwide. Authors said that
women are badly affected by this disease. Authors have de-
picted different types, symptoms, diagnosis and the treatment
for diabetes. Authors showed that use of decision tree and naive
bayes assist in diagnosis of diabetes. The deviation between
the experimental and expected outcome is totally insignificance
with of decision tree and naive bayes data mining techniques.
Rajesh and Sangeetha [25] have explained the role of data
mining techniques in diagnoses of diabetes. Authors stated that
the success of medical science is heavily depended upon the
reliable prediction of different parameters. Authors have taken
different parameters viz. number of pregnancy, diastolic blood
pressure, skin fold thickness, BMI, diabetic pedigree function,
age etc. to probe the performance of the classification algo-
rithms in diagnosing the diabetes. Authors have compared dif-
ferent classification algorithm viz. C-RT, CS-RT, C4.5, ID3,
LDA, SVM, naive bayes etc. From the confusion matrix of the
dataset, authors obtained 91% accuracy rate with C4.5. Santi
Wulan et al. [40] have presented an innovative smooth SVM
solution for diagnosis of diabetes. Authors have used modified
spline. By using modified spline SVM, authors achieved the ac-
curacy rate up to 96.48%.
Santhanam et al. [41] have used hybrid idea of genetic
algorithm and SVM. With this combination, authors achieve
the diagnosis diabetes precision rate of 98.79%. Authors have
tested 50 different cases to unearth the performance of the pro-
jected classification approach. Nilesh Jagdish Vispute et al. [42]
made an empirical assessment of diverse classification tech-
niques used for analysis of diabetic. Authors have used WEKA
tool for their experimentation. 10 cross validation approach
has been used. Authors have compared various techniques viz.
naive bayes, J48 Tree, SMO, RepTree and Random tree. Au-
thors found naïve bayes as best classifier. The maximum rate
of accuracy achieved is 76.03%. Peter [43] stated that diabetes
mellitus is one of persistent and dominant type of disease. Hy-
perglycemia is one of the major reasons behind it. Author has
presented an assessment of diverse algorithms which are used to
mine this disease. Authors have compared the features of NN,
SVM and hybrid approach based upon different performance
metric viz. classification accuracy and processing time required
to predict. Authors concluded that with hybrid approach of
data mining the classification accuracy can be achieved up to
96.65%. Velide Phani Kumar et al. [44] have developed a pre-
dictive model for diagnosis and treatment of diabetes. Authors
have compared different classification techniques. The experi-
ments were carried out on nine different parameters. Authors
found J48 as best classifier as it gives 100% accuracy in their
data set. Authors also examined the time taken in the classifi-
cation process by J48 outperforms other the time consumed by
other techniques.
Ebenezer Obaloluwa Olaniyi et al. [45] have spotted dia-
betes using ANN. Authors stated that a person suffering from
diabetes is normally affected with problems like frequent uri-
nation, polyphagia, polydipsia etc. Authors do experimentation
with 500 training sample and 268 testing samples. Authors
compared various techniques viz. BSS, KNN, C4.5 and BPNN
(Back propagation neural network). During experimentation,
311
authors concluded that BPNN is able to attain 82% accuracy
and is higher than other said techniques. Aljumah et al. [46]
has performed a study on Saudi Arabians. Authors found that
diabetes is significantly increasing the Saudi Arabian people.
Author performed study on both young and old people. As per
their study, the order of treatment is different for young and
old patients. Authors have used one of the data mining tools of
Oracle Corporation. Authors performed rigorous study on the
basis of different parameters viz. drug, diet, weight, smoking,
exercise and insulin. Authors concluded that drug treatment is
effective for both young and old diabetes patients. However, it is
dominant in older. Authors further revealed that except smoking
and insulin, the prediction of effectiveness in treatment using
diet control, weight control and using exercise is more in older
than young’s.
Marinov, Mosa et al. [47] have collected data from MED-
LINE database. Authors examined the role of data mining in
diabetes research. As per authors, data mining is very benefi-
cial for diabetes research. Moreover, it will improve the health
status of diabetic persons more effectively and efficiently.
S. Nagarajan et al. [48] used four algorithms ID3, naive
bayes, C4.5 and Random tree to diagnose Gestational Dia-
betes in pregnant women. Authors mined the data set of 600
instances. Each instance has six attributes. The study has publi-
cized that the Random Tree with error rate 0.000 (lowest among
all) and accuracy value with 0.938 (highest among all) act as the
best algorithm. K.R. Lakshmi et al. [49] performed the com-
parison of ten dissimilar data mining algorithms – C4.5, SVM,
k-NN, PNN, BLR, MLR, PLS-DA, PLS-LDA, k-means and
Apriori to get best results for diagnosing diabetes in patients.
Authors found that PLS-DA give optimal results as compared
to other techniques. The rate of accuracy achieved with PLS-
DAis 74%.
6.3. Diagnosis of heart problems using data mining
approaches
Cardiovascular disease is a most precarious disease. There is
a variety of circumstance that influences the heart of a person.
As per WHO record, more than ten million casualties occur due
to the heart related problems. Indian are significantly affected
by this disease. The problem related to blood vessels, coronary
artery, abnormal heart rhythm and stroke are dominated heart
related diseases. Data mining method or techniques has been
used to predict the risk of cardiovascular disease so that one
can detect it on early stage and have adequate treatment.
Ilayaraja and Meyyappan [20] have developed a method to
predict the risk of heart disease using frequent itemset. Author
has taken 19 different symptoms which are the reasons behind
heart disease. Authors compared their result with other tech-
niques viz. apriori algorithm, IMSIA algorithm, semi-apriori
algorithm etc. Authors found their algorithm work well as com-
pared to other existing techniques. B. Venkatalakshmi, M.V.
Shivsankar [23] stated untimely diagnose of heart problem may
lead to illness or even death in some cases. Authors have tried
to diagnose the heart disease using two different data mining
techniques viz. naive bayes and decision tree. Authors found
312 M. Sharma et al. / IRBM 38 (2017) 305–324
that naive bayes gives optimal results as compared to decision
tree. The accuracy of naïve bayes is 85.03%. Taneja [24] has de-
signed a predictive model which is used to mine dataset based
upon Transthoracic Echocardiography Report. Author has col-
lected data from one of the reputed hospital of Chandigarh.
Three different supervised machine learning techniques viz.
J48 classifier, multilayer perception and naive bayes have been
used. The accuracy of classification obtained was 95.56.
Soni et al. [26] have provided an overview of data min-
ing techniques used in prediction heart diseases. Authors have
taken 15 different parameters for their study. Authors have stud-
ied the working and performance of KNN (k-nearest Neural
Network), naïve bayes and Decision Lists. Authors concluded
that the performance of decision tree is best. The rate of ac-
curacy is 99%. Moreover, in some circumstances naïve bayes
gives a solution very close to as given by the decision tree.
Masethe and Masethe [50] have tried to predict heart disease
problem. Authors have compared and contrasted the perfor-
mance of five different classification algorithms i.e. J48, Rep-
Tree, naïve bayes, CART and bayes net. Authors also used
different evaluation criterion like kappa statistics, mean abso-
lute error, root mean square value, relative absolute error and
root relative squared error. Authors attained 99% accuracy with
J48.
Dangare an Apte [51] stated that data mining techniques are
very useful in predicting the heart disease based upon the his-
tory of the patient. Authors stated that normally, 13 attributes
like sex, blood pressure, cholesterol etc. are used to diagnose
the situation. Authors extended their study by adding two more
parameters viz. obesity and smoking. Authors used and com-
pared the performance of different data mining techniques in
predicting the status of heart disease. Authors found that neural
network gives 100% accuracy in determining the heart disease
status. Sudha et al. [52] stated that stroke is life threatening dis-
ease and is one of the major causes behind the deaths. Authors
tried to analyze the data of stroke patients. They used different
data mining techniques viz. neural network, naïve bayes and
decision tree. Author got better results with the neural network.
Authors proposed a model that predicts the nature of data based
upon the feature subset selection. The accuracy of proposed
framework is 91%. Panzarasa et al. [53] have used data min-
ing techniques to the stroke register in one particular area of
Italy. Authors analyzed around 5000 different cases. The study
was carried out to analyze the delay between the stroke onset
and the hospital admission. Authors used classification tree to
mine the data. Though, they get both expected and unexpected
results. G. Subbalakshmi et al. [54] developed a decision sup-
port for heart related diagnostic system by using naïve bayes
data mining technique. They had taken basic parameter like age,
gender, blood pressure etc. to foresee the likelihood of getting
heart disease.
V. Krishnaiah et al. [55] have developed a fuzzy KNN based
prophetic system for heart diseases. Authors claimed that num-
ber of other researchers has merely used the in intrinsic clas-
sifier of the mining software for diagnosis or forecasting the
data related to heart patients and ignored the data uncertainty.
Therefore, authors used fuzzy approach to remove the uncer-
tainty and to improve the accuracy level of prediction related to
heart patients. Author worked on a data set consisting of 550
records. Each record has 13 different parameters. Saba Bashir
et al. [56] have proposed an innovative ensemble classifier
based upon five different techniques viz. naïve bayes, decision
tree induction based on GINI index, information gain, mem-
ory based learner and SVM. Authors approach (MV5) found
to be an improve one. The rate of forecasting accuracy, sensi-
tivity and specificity attained using MV5 are 88.52%, 86.96%
and 90.83% respectively. Nishara and Gomathy [57] have used
a hybrid data mining technique to diagnose heart disease. Au-
thors have proposed a hybrid model by combining classification
(C4.5), clustering (k-means) and association (maximal frequent
itemset) and found it as a more promising as compared to other
techniques. The rate of precision, recall and accuracy achieve
with proposed model are 0.82, 0.89 and 89% respectively. Mai
Shouman et al. [58] have used k-NN in diagnosis of heart re-
lated problems. Authors used standardized data set from Cleve-
land Clinic Foundation benchmark. There are around 300 rows
in the experimental data set. Authors varied the value of k from
1 to 13. The rate of accuracy varied in between 94% to 97.1%.
In addition, authors observed that voting does not improve the
performance of k-NN in forecasting the heart related problems.
6.4. Role of data mining in examining other diseases
Beside diabetes and cardiovascular disease, there exist num-
ber of other problems that affects the life and system of a per-
son. The remaining part of this sub-section depicts the role of
data mining in diseases other than diabetes and cardiovascular.
Haofan Yang et al. [22] have designed a framework to di-
agnose the lung cancer. Authors designed a framework that
find the association between the clinical and pathological in-
formation related to lung cancer patients. Author used Apriori
algorithm to train the system. Gouda I. Salama et al. [37] have
explored the working of individual and fusion of classifier in
breast cancer diagnosis. Authors found that the best rate of clas-
sification is achieved with four fusion classifier i.e. when four
different type of classifier are blended together.
Ling Chen et al. [59] have developed a model called MyPHI
that is used to compute the personal health index of a person.
With MyPHI, one is able to know the status of their health. Au-
thor found that their system is better than other benchmark of
healthcare industry. Swati Gupta [60] tried to diagnose one of
the eye disease i.e. diabetic retinopathy. Authors have exam-
ined the features viz. micro aneurysm and exudates. Authors
found better results with SVM and the rate of accuracy re-
alized is 86%. Authors compared results of SVM and KNN.
Lincoln F. Silva et al. [61] have developed a predictive sys-
tem for diagnosis of breast cancer. Authors worked on the pa-
tients data of hospital of Antonio Pedro University. Authors
used Bayesian, neural network and decision tree to classify
the patients suffering from breast cancer. Authors found that
the performance of both neural network and decision tree is
same and the rate of accuracy is 90.91. The data and training
set used for experimentation was very small in size. Chaitali
et al. [62] have compared the results of different classifica-

Table 3
Survey of data mining techniques.
Authors and year of
publication
Gwenolé Quellec, Mathieu
Lamard, Ali Erginay, Agnès
Chabouis, Pascale Massin,
Béatrice Cochener, Guy
Cazuguel [68]
Ilayaraja M., Meyyappan [20]
Aiswarya Iyer, Jeylatha and
Ronak Sumlbaly [21]
Haofan Yang, Yi-Ping Phoebe
Chen [22]
Chaitali Vaghela, Nikita Bhatt,
Darshana Mistry [62]
Masethe and Masethe [50]
S. Peter [43]
Kavita Chaudhary, Pinki Bajaj
[69]
Aljumah, Ahmad et al. [46]
Taneja [24]
Dangare and Apte [51]
Sudha, Gayathri et al. [52]
K. Rajesh and V. Sangeetha
[25]
M. Sharma et al. / IRBM 38 (2017) 305–324 313
Type of Methodology and tool Results
disease
Eyes Apriori Authors have successfully used mining techniques to diagnose eyes
pathological problems.
Heart Apriori algorithm, IMSIA, A predictive method is developed to predict the risk of heart problems.
Semi-priori algorithm
Diabetes J48, Naïve Bayes Diabetes has affected millions of people over the globe. Authors have
applied classification algorithm to data collected from pregnant women.
Authors have computed the confusion matrix using WEKA tool.
Experimental results that the results of naïve bayes outperformed the
results obtained with decision tree.
Lung Cancer Apriori algorithm Authors have proposed a framework to find association between the clinical
information and pathological information for lung cancer patients. The
framework found to be efficient in lung cancer pathologic Staging.
Breast Naïve bayes, RBFN Authors have compared the results of different classification techniques for
Cancer, (Radial base function the diseases like breast cancer, heart and diabetes. For breast cancer,
Heart network), neural network Authors have tested 286 samples. For breast cancer, rate of accuracy
Diabetes achieved is 75.4%. For heart disease, 303 samples were tested and the
performance of RBFN (Radial base function network) is outperformed. For
diabetes, 768 instances were taken. For diabetes, naïve bayes gives best
performance. The rate of accuracy attained 75.39%.
Heart J48, RepTree, Naïve A comparative study on five different classification algorithms is carried
Bayes, CART and Bayes out to analyze the heart related problems. To evaluate the efficiency of
NET different classification algorithms, several parameters viz. kappa statistics,
mean absolute error, root mean square values, relative absolute error and
root relative squared error are also computed.
Diabetes Neural Network, Support A survey is conducted on different classification techniques used to
Vector Machine, Hybrid diagnose diabetes. Authors found the hybrid approach predicts more
approach accurately as compared to other classification approaches.
RCT Cross validation and Authors have used classification technique to diagnose the need of root
Decision Tree canal treatment. It was found that by using certain metrics like age,
brushing, smoking, tooth decay etc., one is able to predict the need of RCT
by using classification techniques of data mining.
Diabetes Oracle Data Corporation Authors carried out meticulous study on the different types of treatment
suggested for both young and older. As per the study, except smoking and
insulin, prediction of effectiveness in treatment using weight, diet, exercise
and drug is more in older than young’s.
Heart Decision Tree Author has collected analyzed the data collected from PGI, Chandigarh,
Classification, Bayesian India. WEKA tool has been used to develop a predictive model for heart
Classifier and Neural disease. Author found that optimal results with J48. 95.56% accuracy has
Network been achieved
Authors have used 15 different parameters rather than 13 for predicting
heart disease. Authors have taken 571 records out of which 303 are used for
training and rest are used for testing. Authors found that for their particular
set of data, neural network gives 100% accurate results.
Stroke Decision Tree, Naïve Stroke is one of the most dangerous diseases that may lead to death or life
Bayes, Neural network time disability. Authors tried to predict the chances of stroke by measuring
certain parameters. It was found the neural network gives better
performance as compared to decision tree and naïve bayes.
Diabetes Not specified Authors have performed a comparative study of different classification
algorithms in diagnosis the diabetes. Authors found that C4.5 classification
algorithm gives better results efficiency. Authors also suggested that one
can further improve the design of C4.5 to achieve better results.
(continued on next page)
314 M. Sharma et al. / IRBM 38 (2017) 305–324
Table 3 (continued)
Authors and year of Type of Methodology and tool
publication disease
Soni, Ansari et al. [26] Heart Decision Tree, Bayesian
Network, KNN, Neural
Network, Genetic
Algorithm
Marinov, Mosa et al. [57] Diabetes
Subbalakshmi, Kumar et al. Heart Naïve Bayes
[54] disease
Silvia Panzarasa, Silvana Stroke Not specified
Quanglini, Lucia Sacchi,
Anna Cavallini, Giuseppe
Micieli and Mario Stefanelli
[53]
tion techniques for the diseases like breast cancer, heart and
diabetes. For breast cancer, Authors have tested 286 samples.
For breast cancer authors observed the 75.4 diagnose accu-
racy. For heart disease 303 samples were tested and the per-
formance of RBFN (Radial base function network) is outper-
formed. For diabetes, 768 instances were taken. For diabetes,
Naïve Bayes gives best performance. The rate of accuracy at-
tained 75.39%. P. Ramachandran et al. [63] developed a hybrid
model for early detection and prevention of cancer. Authors
have used a hybrid approach of clustering and decision tree.
Authors evaluated the performance of their model by using
support vector machine. The evaluation has been done based
upon three different statistical parameters viz. accuracy, sensi-
tivity and specificity. The cancer detection is based upon several
parameters like age, education, living area, anemia, history of
family, weight loss and several habits like smoking, alcohol,
chewing etc.
Hayrettin Evirgen et al. [64] have developed a predictive sys-
tem for diagnosis of retinopathy. Retinopathy is one of the eye
problems that occur mainly in the people who are affected with
diabetes for a significant period of time. Authors proved that
with the use of naïve bayes, one is able to achieve 89% of ac-
curacy in diagnosis of retinopathy. A. Kalairasai et al. [65] has
developed a framework for detecting oral cancer. Authors have
experimented with NMDS database. Authors have compared
different techniques viz. apriori algorithm, C4.5, SVM, naïve
bayes and random forest. In the experimental data, random for-
est gives best possible results with NMDS dataset. The rate of
accuracy attained is 83%. Author proposed that a hybrid com-
bination of C4.5 and random forest may further enhance the
accuracy level of the predictive model. Mohanty et al. [66] have
developed a classification model for detecting breast cancer.
Authors have taken 26 parameters and found them more suit-
able for detecting the cancer. Association rule mining is used
to classify normal and effected cases. Shweta Khraya [67] has
Results
Authors performed experimentation for predicting the problems related to
heart diseases. Authors found the decision tree outperforms other
classification techniques. The rate of accuracy achieved using naïve bayes,
decision tree and clustering is 96.5%, 99.2% and 88.3% respectively.
Authors have performed a systematic review of used data mining
techniques for diabetes. Author has collected 31 different papers indexed in
Medline. Authors stated that one should analyze the genomic data as
diabetes is a genetic disease.
Authors developed a DSS for predicting the risk factor in heart disease
using naïve bayes. Authors have taken basic parameters for developing the
system.
Authors used data mining techniques determine the effect of
non-compliance to clinical guidelines proposed for stroke patients. The
study is confined to one of the region of Italy. Authors have analyzed
Stroke Unit Network (SUN) register. The objective of the study was to
analyze the delay between the stroke onset and the hospital admission.
Authors used classification tree to mine the data. Though, they get both
expected and unexpected results.
Fig. 3. Accuracy in diagnosis of diabetes.
reviewed a literature on the diagnosis and prospects of breast
cancer. Author found decision tree as an optimal predictor. Rate
of accuracy achieved is 93.62%. Author suggested to develop a
web based application so that the people from the remote area
can self-diagnose their status based upon the concerned param-
eters.
Table 3 highlight the crux of the above mentioned research
works.The performance of data mining techniques strongly de-
pends upon the selected parameters, training data and the test-
ing data set.
Table 4 represents the accuracy and the data mining ap-
proaches of some of key researchers used to diagnose different
types of diseases.
Fig. 3 represents the level of accuracy achieved using dif-
ferent classification techniques for spotting diabetes. It is re-
vealed that depending upon selection of parameters and data
set used, the prediction accuracy can be achieved between 75 to
100%.
Fig. 4 represents the level of accuracy achieved using dif-
ferent classification techniques for spotting heart diseases. It is
 M. Sharma et al. / IRBM 38 (2017) 305–324 315

Table 4
Accuracy level of used data mining techniques.
Authors and year of publication Type of disease Accuracy Best approach
Aiswarya Iyer, Jeylatha and Ronak Sumlbaly [21] Diabetes 79.56 Naïve bayes
Masethe and Masethe [50] Heart 99 J48
S. Peter (2014) [43] Diabetes 96.68 Hybrid approach
Kavita Chaudhary, Pinki Bajaj (2014) [69] RCT 93.06 Decision Tree with Cross Validation
Taneja (2013) [24] Heart 95.56 J4.8
Sudha, Gayathri et al. (2012) [52] Stroke 91 Neural network
K. Rajesh and V. Sangeetha (2012) [25] Diabetes 91 C4.5
Jyoti Soni et al. (2011) [26] Heart 99.2 Decision Tree
Silvia Panzarasa, Silvana Quanglini, Lucia Sacchi, Stroke 99 Neural network
Anna Cavallini, Giuseppe Micieli and Mario
Stefanelli (2010) [53]
Chaitrali S Dangare, Sulabha S. Apte (2012) [51] Heart disease 100 Neural network
B. Venkatalakshmi, M.V. Shivsankar (2014) [23] Heart disease 85.3 Naïve bayes
A. Kalairasai (2013) [65] Oral Cancer 83 Random forest
Shweta Khraya (2012) [67] Breast Cancer 93.62 Decision tree
Lincoln F. Silva, Giomar O. Sequeiros, Maria Lúcia Breast Cancer 90.91 Decision Tree and Neural Network
O. Santo, Cristina A. P. Fontes, Débora C.
Muchaluat-Saadeand Aura Conci (2015) [61]
Santi Wulan Purnami, Jasni Mohamad Zain, Diabetes 96.58 Smooth SVM
Abdullah Embong (2010) [40]
T. Santhanam, M.S. Padmavathi (2015) [41] Diabetes 98.79 GASVM
Swati Gupta, Karandikar AM (2015) [60] Ophthalmology 86 SVM
Velide Phani Kumar (2014) [44] Diabetes 100 J48
Ebenezer Obaloluwa Olaniyi, Khashman Adnan Diabetes 82 BPNN
(2014) [45]
Chaitali Vaghela, Nikita Bhatt, Darshana Mistry Breast Cancer 75.4 Decision tree
(2015) [62]
Chaitali Vaghela, Nikita Bhatt, Darshana Mistry Diabetes 75.39 Naïve Bayes
(2015) [62]
Chaitali Vaghela, Nikita Bhatt, Darshana Mistry Heart Disease 83.2 RBFN
(2015) [62]
Mahua Nandy (2013) [70] Breast Cancer 96.7 SVM
M.A. Nishara Banu and B. Gomathy (2014) [57] Heart Disease 89 Hybrid
Mai Shouman et al. (2012) [58] Heart Disease 97.1 k-NN
P. Rajeswari, G. Sophia Reena (2010) [36] Liver disorder 96.52 Naïve bayes
P. Rajeswari, G. Sophia Reena (2010) [36] Liver disorder 97.10 FT Tree
P. Rajeswari, G. Sophia Reena (2010) [36] Liver disorder 83.4 K-Star

data set grows significantly. It is also found that the accuracy in

diagnosis of chronic diseases namely heart and diabetes with
J48 and decision tree respectively is above par as compared
to other classification approaches. In addition, the performance
of J48 and decision tree is different for diagnosis of other dis-
eases.

7. Effect of preprocessing on predictive accuracy of

lifestyle based disease diagnosis

Before mining, the data collected from different sources like

Fig. 4. Accuracy in diagnosis of heart disease.
revealed that depending upon selection of parameters and data
set used, the prediction accuracy can be achieved between 85 to
100%.
From Table 3, it is observed that authors have achieved the
100% diagnosis accuracy in heart and diabetes diseases. In ad-
dition, independent of size of testing and training data, the rate
of accuracy in diagnosis of different diseases lies between 70
to 100%. However, it would be unrealistic if the size of testing
signal, images or reports has to be cleaned and normalized to
avoid any noise or data variation. The patient’s data are mo-
mentously prone to missing, inconsistent and noisy data as it
is collected from different sources. The data inconsistency will
affect the predictive rate of disease diagnosis accuracy. Data
preprocessing assist to improve the quality of healthcare data.
Data cleaning and data reduction are two major categories of
data preprocessing techniques. Data cleaning helps in noise re-
moval and to minimize the data inconsistency. Data reduction
are useful in reducing data redundancy, summarizing and in fea-
316 M. Sharma et al. / IRBM 38 (2017) 305–324
ture selection process. Researchers have found that the use of
data cleaning and dimension reduction gives better mining re-
sults (Table 5).
The initial data set of patients may suffer from several
anomalies. For better predictions, the data should be cleaned
by using different methods of preprocessing. Amir R. Razavi et
al. [71] have tried to analyze the effect of preprocessing on the
breast cancer patients dataset. Authors cleaned the data set of
3949 patients by using different laws of logic. Missing values
were manipulated using expectation maximizing techniques.
Moreover, CCA (canonical correlation analysis) has been used
to reduce data dimension. Authors observed that use of expec-
tation maximization and CCA has improved the classification
process. The rate of accuracy achieved without preprocessing,
after replacing missing values and after dimension reduction
using DTI is 54%, 57% and 67% respectively.
Soni and Ansari [26] have used textual and numeric based
Cleveland heart disease dataset. Authors used K-means clus-
tering to extract the homogeneous data. GA is used for feature
selection so that optimal and normalized data set can be con-
structed.
R. Ramani et al. [27] have used four different types of filters
viz. median, adaptive median, mean and wiener to deal with
noisy images. Authors examined the performance of different
filters in context to the three different types of noise Gaussian,
salt & pepper and speckle. It was observed that results produced
by adaptive median filter are better than other filters.
Jianfeng Zhang et al. [29] have diagnosed diabetes using
SVM. GA and PCA are used for data preprocessing. Authors
have used tongue image for diabetes diagnosis. GA is used for
feature selection. PCA is used to minimize the magnitude of at-
tribute’s range so that the scale from −1 to +1 can be set for all
selected attributes.
Gouda I. Salama et al. [37] diagnosed breast cancer using
UCI data sets. Author stated that the use of PCA improved the
classification rate of the fusion of J48 and SMO. The rate of
accuracy achieved using above mentioned classifier fusion and
feature reduction technique (PCA) is 97.56%.
Rajesh Kumar, Rajeev Srivastava, and Subodh Srivastava
[30] have detect cancer using biopsy images. Authors have
used preprocessing, segmentation and feature extraction tech-
niques to get more precise diagnosis results. The enhancement
techniques were used to remove staining and to handle wicked
contrast. The study were performed over four different types
of tissues viz. connective, muscular, nervous and epithelial.
Moreover, different feature extraction techniques like wavelet,
HOG, LTE, texture and shape were used. Authors found that
the combination of above mentioned preprocessing techniques
with KNN is able to provide 92.19% rate of accuracy.
Razieh Asgarnezhad et al. [72] have discussed different pre-
processing techniques used to handle missing values for dia-
betes data set. In addition, different methods for attributes se-
lection are also concisely stated. Authors explored the usage
of backward elimination and forward selection, brute force and
genetic algorithms for choosing the subset of attributes. Ad-
ditionally, the effect of replacing missing values with mean,
median and k-NN approach was also examined. It was found
that the use of SVM without any preprocessing techniques was
able to achieve 72.04% predictive rate of classification. It was
observed that the use of preprocessing techniques (attribute se-
lection using GA, Replacing missing value with mean) signif-
icantly raised the classification rate. The rate of classification
achieved using SVM after preprocessing is 83.33%.
Ashok Kumar and Govidsamy [73] have compared the per-
formance of five different classification algorithms viz. SVM.
Naïve bayes, regression, bayes net and decision for diabetes
patients. Authors also examine the effect of greedy stepwise
preprocessing technique on the predictive rate of classification.
It was observed that after applying greedy stepwise features
selection technique the performance of each classification tech-
nique has been improved.
Jayalskshmi and Santhakumaran [74] have examined the im-
pact of preprocessing techniques on the performance of ANN
in diagnosing diabetes patients. Authors found the use of PCA
with ANN and replacing missing value with median achieved
better accuracy as compared to other preprocessing techniques
like min–max, Z-score and Mapstd.
H. Hamidi and A. Daraei [75] have explored the effect of
pre and post processing on heart diagnosis problems. Authors
surveyed 43 different papers on heart diagnosis using data min-
ing techniques. They found that features selection is one of the
common preprocessing techniques. In addition, data transfor-
mation also play important role in improving the performance
of diagnosis process.
Hongmei Yan, Jun Zheng, Yingtao Jiang et al. [76] have im-
plemented real coded GA to determine the best possible and
critical features for heart disease diagnosis. Authors extracted
24 features out of 40 and stated that these 24 features are suffi-
cient to precisely diagnose any heart disease problem.
8. Relationship of data type and data mining algorithm
Researchers have used different data sets to diagnose dif-
ferent lifestyle based human disorders. Some of the common
datasets used for diabetes diagnosis are PIMA, UCI and Hos-
pital’s data. Most of the researchers have used textual and nu-
meric types of attributes in diabetes diagnosis. However, some
of the researchers have also found effect results by using tongue
images.
From Table 6, following observations have been made:
– For diabetes diagnosis, the best results are achieved when
the fusion of GA and SVM are applied on text and numeric
based datasets.
– Tongue images are also seems to be useful in diabetes di-
agnosis. However, rate of accuracy achieved using tongue
images is not as high as accomplished with text and nu-
meric based datasets.
– Like diabetes, cardio human disorders are also well diag-
nosed with text and numeric data sets. However, authors
have also used images, signal and sound for heart dis-
ease diagnosis. Decision tree with text and numeric based
datasets is found to be more effective for cardio problem
diagnosis.
 M. Sharma et al. / IRBM 38 (2017) 305–324 317

Table 5
Effect of preprocessing on classification accuracy.
Authors and yea Preprocessing technique Classification Accuracy Accuracy
techniques before after
preprocessing preprocessing
Kumar and Govindsamy (2015) [73] Greedy Stepwise SVM 77.47% 77.73%
Naïve bayes 76.30% 77.60%
BayesNet 78.25% 78.25%
Decision Tre 77.60% 79.81%
Regression 77.34% 77.60%

Razieh Asgarnezhad et al. (2017) [72] Backward Elimination & Forward Selection, SVM 72.04 74.58
Replacing missing value with mean
Backward Elimination & Forward Selection, SVM 72.04 77.78
Replacing missing value with Median
Backward Elimination & Forward Selection, SVM 72.04 74.14
Replacing missing value with k-NN
Brute Force, Replacing missing value with Mean SVM 72.04 75
Brute Force, Replacing missing value with Median SVM 72.04 79.03
Brute Force, Replacing missing value with k-NN SVM 72.04 77.55
GA, Replacing missing value with Mean SVM 72.04 83.33
GA, Replacing missing value with Median SVM 72.04 80.36
GA, Replacing missing value with k-NN SVM 72.04 76.56

Jayalskshmi and Santhakumaran Min–max, omitting missing value ANN – 71.59

(2010) [74] Z-score, omitting missing value ANN – 76.38
Mapstd, omitting missing value ANN – 68.35
PCA, omitting missing value ANN – 67.84
Min–max, replacing with zero ANN – 68.49
Z-score, replacing with zero ANN – 72.47
Mapstd, replacing with zero ANN – 69.27
PCA, replacing with zero ANN – 66.67
Min–max, replacing with Mean ANN – 71.36
Z-score, replacing with Mean ANN – 69.87
Mapstd, replacing with Mean ANN – 67.25
PCA, replacing with Mean ANN – 99.9
Min–max, replacing with k-NN ANN – 69.87
Z-score, replacing with k-NN ANN – 68.09
Mapstd, replacing with k-NN ANN – 60.77
PCA, replacing with k-NN ANN – 99.8

Jianfeng Zhang et al. (2017) [29] GA and PCA SVM 77.83 78.77
Rajesh Kumar, Rajeev Srivastava, and Enhancement, HOG, LTE, Wavelet, Shape, KNN – 92.19
Subodh Srivastava (2015) [30] Texture
K. Vembandasamy, T. Karthikeyan Outlier detection FCM 84 93
(2016) [77]
K. Vimala and Dr. V. Kalaivani (2013) [31] DWT ANN – 97.68

Gouda I. Salama et al. (2012) [37] PCA Fusion of J48 and SMO 96.28 96.99

– For breast cancer, images are seem to be more effective.
SVM provides better results in early diagnosis of breast
cancer.
– For digestive disorders different types of data viz. text, nu-
meric and images have been used. For liver disorder identi-
fication, text and numeric based dataset with C4.5 achieved
better classification results.
9. Analysis of research publications using Google scholar
The important facet of this paper is to emphasize the medical
sciences domain where already lots of data have been mined
and the areas that still require an attention for data mining. For
this, last seven years of data has been extracted from Google
Scholar’s. Tables 7, 8, 9, 10 and 11 represent the status of use
of preprocessing techniques in healthcare diagnosis system for
diabetes, cardio, digestive, ophthalmology and dentistry human
disorders.
From Fig. 5, the rate of usage of preprocessing in diabetes
diagnosis lies between 8.48 to 14.80%. From last seven years,
the rate of usage of preprocessing is diabetes diagnosis is in-
creased.
Like diabetes, the rate of usage of preprocessing in cardio
disorder has been also increased in last seven years. The max-
imum rate of usage of preprocessing in cardio disorder was in
year 2017 and rate of usage was 17.75% (Fig. 6).
318 M. Sharma et al. / IRBM 38 (2017) 305–324

Table 6
Relationship between disease, datatype and accuracy of data mining approach.
Disease Data type Techniques Accuracy
Diabetes (Aiswarya Iyer, Jeylatha and Ronak Sumlbaly) [21] Text and numeric Naïve bayes 79.56
Diabetes (T. Santhanam, M.S. Padmavathi) [41] GA-SVM 98.79
Diabetes (Mahmoud Heydari, Mehdi Teimouri, Numeric and nominal ANN 97.44
Zainabolhoda) [78]
Diabetes (S. Peter) [43] Text and numeric Neural Network, Support 96.68
Vector Machine, Hybrid
approach
Diabetes (K. Rajesh and V. Sangeetha) [25] Textual and numeric C4.5 91
Diabetes (Velide Phani Kumar) [44] Text and numeric J48 100
Heart (Masethe and Masethe) [50] Nominal J48, RepTree, Naïve 99
Bayes, CART and Bayes
NET
Diabetes (Santi Wulan) [40] Integer, Real Smooth SVM 96.58
Diabetes (Nilesh Jagdish Vispute et al.) [42] Text and numeric Naïve Bayes 76.03
Diabetes (Jianfeng Zhang et al.) [29] Tongue Images SVM 83.06
Diabetes (N.V. Cibin, S. Wilfred Franklin, N.V. Ajin) [79] Tongue Images SVM 66.26
Diabetes (N.V. Cibin, S. Wilfred Franklin, N.V. Ajin) [79] Tongue Images LCA 85.52
Heart Disease (Santi Wulan) [40] Integer, Real Smooth SVM 94.15
Heart (Soni and Ansari) [26] Text and numeric Decision Tree 99.2
Heart diseases (M.A. NisharaBanu and B. Gomathy) [57] Text and numeric Hybrid 89
Heart disease (B. Venkatalakshmi, M.V. Shivsankar) [23] Naïve bayes 85.3
Heart (Taneja) [24] Transthoracic J48 classifier, multilayer 95.56
Echocardiography perception and naive
Report bayes
Heart (Sujata Joshi and Mydhili K. Nair) [11] Text and numeric Decision Trees 92.2
Heart disease (Hyeongsoo Kim et al.) [83] Ultrasound Neural Network 86.24
images and elec- Bayesnet 79.65
trocardiogram C4.5 83.63
signal SVM 89.51
CMAR 89.46
Kumar and Govindsamy [73] Text and numeric Decision Tree 79.81
Breast cancer (Ramani) [27] Images Not mentioned Not mentioned
Breast cancer (Shweta) [67] Images Decision tree 93.62
Breast cancer (Chaitali) [62] Images Naïve bayes, RBFN 75.4
(Radial base function
network), neural network
Breast cancer (Mahua Nandy) [70] Images SVM 96.7
Cancer (Rajesh Kumar, Rajeev Srivastava, and Subodh) [30] Biopsy Images KNN 92.19
Liver disorder (P. Rajeswari, G. Sophia Reena) [36] Text and numeric FT Tree 97.10
Ophthalmology (Swati Gupta, Karandikar AM) (2015) [60] Images SVM 86
Liver disorder (A.S. Aneeshkumar, C. Jothi Text and numeric C4.5 99.20
Venkateswaran) [39]

Table 7 Table 8
Year wise publication details with and without preprocessing techniques. Year wise publication details with and without preprocessing techniques.
Year Article on diabetes Article on diabetes Year Article on cardio Article on cardio disorder
diagnosis using diagnosis using data disorder diagnosis diagnosis using data
data mining mining with preprocessing using data mining mining with preprocessing
2011 6900 585 2011 13000 1480
2012 8380 735 2012 13900 1480
2013 8920 1090 2013 14900 1910
2014 8790 1110 2014 13000 2000
2015 9100 1230 2015 14900 2220
2016 9790 1390 2016 15100 2460
2017 5980 885 2017 8280 1470

From Table 10 and Fig. 8, it is observed that researchers have

From Table 9 and Fig. 7, it is observed that rate of usage of almost ignored the data mining in ophthalmology disorder di-
preprocessing in digestive disorder diagnosis is very low. The agnosis. However, the rage of usage of preprocessing has been
rate of digestive disorder diagnosis lies between 4.58 to 8.93%. increased in last seven years.
 M. Sharma et al. / IRBM 38 (2017) 305–324 319

Table 9 Table 11
Year wise publication details with and without preprocessing techniques. Year wise publication details with and without preprocessing techniques.
Year Article on digestive Article on digestive Year Article on dentistry Article on dentistry
disorder diagnosis disorder diagnosis using disorder diagnosis disorder diagnosis using
using data mining data mining with using data mining data mining with
preprocessing preprocessing
2011 5850 268 2011 3610 107
2012 5790 315 2012 3210 136
2013 6260 559 2013 3840 297
2014 6060 387 2014 3180 167
2015 6070 461 2015 3250 176
2016 6350 473 2016 2860 178
2017 3980 259 2017 1940 141

10. Practice connotation

Table 10
Year wise publication details with and without preprocessing techniques.
It was found that data mining techniques were effectively
Year Article on Article on ophthalmology used in mining various health related problems particularly that
ophthalmology disorder diagnosis using deals with heart, diabetes, stroke, cancer etc. However, majority
disorder diagnosis data mining with
using data mining preprocessing
of the experimentation were based upon some specific data sets.
Researchers have normally focused on certain parameters and
2011 1080 63
2012 1360 80 not all. By improving the data set contents, one can mine more
2013 1500 137 useful results and information.
2014 1240 134
2015 1230 142 11. Need of predictive model for correcting lifestyle based
2016 1210 136
2017 789 102
human disorder

Barbara et al. [80] exposed that a smoker has 9% more

Like ophthalmology, a very small amount of mining work
has been done in dentistry disorder diagnosis. The range of us-
age of preprocessing in dentistry diagnosis lies between 2.96 to
7.73% (Fig. 9).
chances of being infected with nuclear cataract (a type of
cataract). Beside, smoking is also one of the major reasons
behind diabetic retinopathy. Similarly, overdose of alcohol con-
sumption (more than four drinks) increases the chances of
cataract by 34%. Moreover, cataract chances are also increased

Fig. 5. Number of article published using preprocessing in diabetes diagnosis.

Fig. 6. Number of article published using preprocessing in cardio disorder diagnosis.

320 M. Sharma et al. / IRBM 38 (2017) 305–324

Fig. 7. Number of article published using preprocessing in digestive disorder diagnosis.

Fig. 8. Number of article published using preprocessing in ophthalmology disorder diagnosis.

Fig. 9. Number of article published using preprocessing in dentistry disorder diagnosis.

the lifestyle habits like smoking, recreational drug and tongue

Fig. 10. Lifestyle based disease diagnosis related article publication summary.
if a person works under UV-B light range. Blue collar persons
are 50% are more prone to trauma (dysfunction of any compo-
nent of eye) than white collars.
Oral health [81,82] which is generally unseen is an impor-
tant factor that contributes to a vigorous lifestyle. Some of
piercings may lead to gum diseases, bleeding, bad breadth, tin-
gling, lumps and sometimes to oral cancer also. Work or social
stress may also lead to serious issues including oral cancer, gum
disease or tooth decay. It may also lead to Temporomandibular
disorder that may badly affect the jaw joints and other oral mus-
cles.
Fig. 10 represents the last seven years article publication sta-
tus in healthcare industry. The rate of work done in diagnosis
different lifestyle based human disorder related to cardio, dia-
betes, digestive, dentistry and ophthalmology disorder are 42%,
26%, 18%, 10% and 4% respectively.
After careful analysis from Google Scholar, it is observed
that a significant work has been done for mining the data related
to the disease viz. heart, and diabetes. Most of the researchers
have tried to mine data related to cardio patients. However, very
little attention has been paid to analyze the data related to Oph-
thalmology, Dentistry and Digestive disorders. Ophthalmology
 M. Sharma et al. / IRBM 38 (2017) 305–324
Fig. 11. Proposed diagnosis model for lifestyle based human disorders.
and Dentistry are almost ignored by data mining researchers.
However, there exist some chronic diseases in these domains
that may lead to blindness, oral, lung cancer and even death to
some cases. Hence, an attention is considered necessary to ex-
plore the usage of data mining in these areas. Therefore, there
is a need to develop an amalgam prophetic model that assists
in early diagnosis of the diseases related to ophthalmology, oral
and digestive system.
Fig. 11 represents the projected architecture for lifestyle
based human disorder diagnosis. The major components of pro-
jected architecture are biosensors, IoT (Internet of Things), pre-
processing and data reduction techniques, a fusion procedure of
soft computing and machine learning techniques, data analysis
procedure, patients and healthcare professionals. Here, an intel-
ligent procedure is designed to collect patient’s data. The data
can be collected manually or through biosensors which are con-
nected through IoT. A preprocessing module is incorporated to
clean, transform and normalize data. Data reduction model is
used to reduce the unwanted features of data sets. A hybrid
predictive system based upon soft computing and data min-
ing based machine learning techniques should be developed.
By varying different permutation of soft computing and data
mining techniques (GA and SVM, GA and Naïve Bayes, ACO
and Decision Tree, PSO and Association Rules etc.) an effec-
tive predictive model can be designed that will assist in early
diagnosis and treatment of lifestyle based diseases. To further
improve the rate of forecasting accuracy, different symptoms
based ontologies related to ophthalmology, dentistry and diges-
tive system can also be incorporated. The use of fuzzy logic
and information theory may also be incorporated to deal with
uncertainty and missing data. The patient’s data is completely
analyzed by data analysis module. The analyzed data will be
321
then communicated through IoT to both patient and health-
care professional so that proper monitoring and care of patient
can be taken out. Finally, the patient response in regard to the
prescribed treatment should be monitored using social media.
Accordingly, the ontologies and training data should be restruc-
tured from time to time.
The novelty lies in the fusion of machine learning, soft com-
puting, fuzzy logic, ontologies, IoT, biosensors and preprocess-
ing techniques.
12. Conclusion
This paper targets to start an effort for designing predic-
tive model for life style based diseases namely ophthalmology,
dentistry and digestive system. As per Google Scholar, the per-
centage of articles published related to cardio, diabetes, diges-
tive, dentistry and ophthalmology disease diagnosis using data
mining in last seven years are 42%, 26%, 18%, 10% and 4%
respectively. This study confirms that already lots of data have
been mined for cardiology and diabetic issues. Time has come
to aware human masses for correcting their life style to avoid
the wide destructive spectrum of digestive based disorders, oph-
thalmology and dentistry. Therefore, a smart and intelligent
solution is required for the same.
First, it is not feasible to incorporate all the papers in any sur-
vey. However, the best one’s are selected for analyzing the role
of classification and predictive techniques in medical databases.
Second, data mining techniques are used to predict human
disorder aforementioned with high precision and stumpy cost.
Data mining is used to reveal novel clinical and medical ac-
quaintances which are assisting to decrease the death rate of
patients. Early human disorder verdict has significant potential
in providing a better medical treatment. Moreover, data mining
322 M. Sharma et al. / IRBM 38 (2017) 305–324
techniques also help in finding the association between the dis-
ease sand the life style of the people. Moreover, based upon size
and nature of training and testing data sets and with same clas-
sification techniques, different authors found different level of
accuracy. In addition, the researchers have normally focused on
common 13 parameters. They have ignored the parameters like
level of stress due to family, work environment, sedentary, life
style, hereditary etc. which are more imperative for cardiology
and diabetes. Therefore, to get more accurate results from the
momentous testing data set, one has to incorporate numerous
parameters and has to statistically explore them.
Third, several types of datasets like textual, numeric, cate-
gorical, sound, images and bio-signals have been used to di-
agnose lifestyle based human disorders. However, most of the
authors have used textual, numeric and categorical variables.
In addition, tongue images are also seems to be useful in di-
abetes diagnosis. However, rate of accuracy achieved using
tongue images is not as high as accomplished with text and nu-
meric based datasets. For diabetes diagnosis, the best results
are achieved when the fusion of GA and SVM are applied on
text and numeric based datasets. For cardiac problem diagnosis,
authors have used different type of datasets like textual, bio-
signals, heart sound, human voice etc. Decision tree with text
and numeric based datasets is found to be more effective for
cardio problem diagnosis. For cancer detection biopsy images
were found to be more useful. SVM provides better results in
early diagnosis of breast cancer. For digestive disorders differ-
ent types of data viz. text, numeric and images have been used.
For liver disorder identification, text and numeric based dataset
with C4.5 achieved better classification results.
As well, the predictive rate of disease diagnosis has been sig-
nificantly improved by using different data preprocessing tech-
niques like data cleaning, transformation, normalization, outlier
detection and dimension reduction. Moreover, the use of fu-
sion classifier with preprocessing techniques gives better results
as compared to the use of individual classifier. However, the
rate of usage of preprocessing in diagnosis of different lifestyle
based human disorder using data mining related to cardio, di-
abetes, digestive, dentistry and ophthalmology lies between
10.65%–17.75%, 8.48%–14.80%, 4.58–8.93%, 2.96%–7.73%
and 5.83%–12.93% respectively.
So, it was found that a significant work has been done
for mining the data related to the disease viz. heart, and dia-
betes.However, very little attention has been paid to analyze the
data for diseases viz. Ophthalmology, Dentistry and Digestive
system etc. Therefore, an attention is needed to explore the us-
age of data mining in these areas.
Finally, the research should be conceded to develop a hybrid
predictive system using soft computing, data mining and infor-
mation theory. Further, to reduce ambiguity and to perk up the
predictive precision, fuzzy logic and symptom based ontologies
can also be incorporated.
Conflict of interest statement
There are no conflict of interests.
References
[1] Du Hongbo. Data mining techniques and applications – an introduction.
Cengage Learning India; 2010.
[2] Liao Shu-Hsien, Chu Pei-Hui, Hsiao Pei-Yuan. Data mining techniques
and applications – a decade review from 2000 to 2011. Expert Syst Appl
2012;39:11303–10.
[3] Han Jiawei, Kamber Micheline, Pei Jian. Data mining concepts and tech-
niques. USA: Morgan Kaufmann Publishers; 2015.
[4] Dunham Margaret H. Data mining introductory and advanced topics. In-
dia: Pearson Publications; 2013.
[5] Pawar Suvarna, Sikchi Smita. An extensive survey on diagnosis of dia-
betes mellitus in healthcare. In: Proceedings of the international confer-
ence on data engineering and communication technology. Advances in
intelligent systems and computing, vol. 468. 2017.
[6] Kaur Shubpreet, Bawa RK. Future trends of data mining in predicting the
various diseases in medical healthcare systems. Int J Energy Inf Commun
2015;6(4):17–34.
[7] Tomar Divya, Aggarwal Sonali. A survey on data mining approaches for
healthcare. Int J Bio-Sci Bio-Technol 2013;5(5):241–66.
[8] Bajaj Punam, Gupta Preeti. Review on heart disease diagnosis based on
data mining techniques. Int. J. Comput. Sci. Manag. Res. 2014;3(5):1596.
[9] Ahmad Parvez, Qamar Saqib, Qasim Syed, Rizvi Afser. Tech-
niques of data mining in healthcare: a review. Int J Comput Appl
2015;120(15):38–50.
[10] Case A, Lubotsky D, Paxson C. Economic status and health in childhood:
the origins of the gradient. Am Econ Rev 2002;92(5):1308–34.
[11] Joshi Sujatha, Nair Mydhili K. Prediction of heart disease using clas-
sification based data mining techniques. Comput Intell Data Mini
2015;2:503–11.
[12] Bhagya Shree SR, Sheshdari HS. Diagnosis of Alzheimer’s disease using
naive bayesian classifier. Neural Comput Appl 2016.
[13] Bhagya Shree SR, Sheshadri HS, Krishna Murali. Diagnosis of
Alzheimer’s disease using rule based approach. Indian J Sci Technol
2016;9(13):1–6.
[14] Sweilam Nasser H, Tharwat AA, Abdel Moniem NK. Support vector ma-
chine for diagnosis cancer disease: a comparative study. Egypt Inform J
2010;11:81–92.
[15] Divya, Chhabra Raman, Kaur Sumit, Ghosh Swagata. Diabetes detection
using artificial neural networks & back-propagation algorithm. Int J Sci
Technol Res 2013;2(1):9–11.
[16] Sharma Manik, Gurvinder Singh, Rajinder Singh, Gurdev Singh. Analy-
sis of DSS queries using entropy based restricted genetic algorithm. Appl
Math Inf Sci 2015;9(5):2599–609.
[17] Sharma Manik, Gurvinder Singh, Rajinder Singh. Design and analysis of
stochastic DSS query optimizers in a distributed database system. Egypt
Inform J 2015. http://dx.doi.org/10.1016/j.eij.2015.10.003.
[18] Jiang Fei, Jiang HuiZhi Yong, et al. Artificial intelligence in healthcare:
past, present and future. Stroke Vasc Neurol 2017;2017:1–13.
[19] https://archive.ics.uci.edu/ml/datasets.html?for-
mat=&task=cla&att=&area=life&numAtt=&nu-
mIns=&type=&sort=nameUp&view=table.
[20] Ilayaraja M, Meyyappan T. Efficient data mining method to predict
the risk of heart disease through frequent itemset. Proc Comput Sci
2015;70:586–92.
[21] Aiswarya Iyer, Jeyalatha S, Sumbaly Ronak. Diagnosis of diabetes using
classification mining techniques. Int J Data Min Knowl Manag Process
2015;5(1):1–14.
[22] Yang Haofan, Chen Yi-Ping Phoebe. Data mining in lung cancer patho-
logic staging diagnosis: correlation between clinical and pathology infor-
mation. Expert Syst Appl 2015;42:6168–76.
[23] Venkatalakshmi B, Shivsankar MV. Heart disease diagnosis using predic-
tive data mining. Int J Innov Res Sci Eng Technol 2014;3(3):1873–7.
[24] Taneja Abhishek. Heart disease prediction system using data mining tech-
niques. Orient J Comput Sci Technol 2013;6(4):457–66.
[25] Rajesh K, Sangeetha V. Application of data mining methods and tech-
niques for diabetes diagnosis. Int J Eng Innov Technol 2012;2(3):224–9.
 M. Sharma et al. / IRBM 38 (2017) 305–324
[26] Soni Jyoti, Ansari Ujma, et al. Predictive data mining for medical di-
agnosis: an overview of heart disease prediction. Int J Comput Appl
2011;17(8):43–8.
[27] Ramani R, Suthanthira Vanitha N, Valarmathy S. The pre-processing tech-
niques for breast cancer detection in mammography images. Int J Image
Graph Signal Process 2013;5:47–54.
[28] Delshi Howsalya Devi R, Indra Devi M. Effective diagnosis of heart dis-
ease using intern quartile range filter and decision tree classifier. 2016.
[29] Zhang Jianfeng, Xu Jiatuo, Hu Xiaajuan, et al. Diagnostic method of di-
abetes based on support vector machine and tongue images. BioMed Res
Int 2017;2017:1–9.
[30] Kumar Rajesh, Srivastava Rajeev, Srivastava Subodh. Detection and
classification of cancer from microscopic biopsy images using clin-
ically significant and biologically interpretable features. J Med Eng
2015;2015:1–14.
[31] Vimala K, Kalaivani V. Classification of cardiac vascular disease from
ECG signals for enhancing modern health scenario. Health Inform - Int
J 2013;2(4):63–72.
[32] Prakash D, Uma Mageshwari T, Prabakaran K, Suguna A. Detection
of heart diseases by mathematical artificial intelligence algorithm using
phonocardiogram signals. Int J Innov Appl Stud 2013;3(1):145–50.
[33] Wu Hang, Kim Sahong, Bae Keunsung. Hidden Markov model with heart
sound signals for identification of heart diseases. In: Proceedings of 20th
international congress on acoustics; 2010.
[34] Pareek Vishakha, Sharma RK. Coronary heart disease detection from
voice analysis. In: IEEE students’ conference on electrical, electronics and
computer science; 2016.
[35] Tsai Du-Yih, Lee Yongbum. Fuzzy-reasoning-based computer-aided di-
agnosis for automated discrimination of myocardial heart disease from
ultrasonic images. Electron. Commun. Jpn., Part 3, Fundam Electron Sci
2002;85(11):1–8.
[36] Rajeswari P, Reena G Sophia. Analysis of liver disorder using data mining
algorithm. Glob J Comput Sci Technol 2010;10(14):48–52.
[37] Salama GI, Abdelhalim MB, Abd-elghanyZeid Magdy. Breast cancer di-
agnosis on three different datasets using multi-classifiers. Int J Comput Inf
Technol 2012;1(1):36–43.
[38] Sharma Disha, Jindal Gagandeep. Identifying lung cancer using im-
age processing techniques. In: International conference on computational
techniques and artificial intelligence; 2011. p. 115–20.
[39] Aneeshkumar AS, Jothi Venkateswaran C. Estimating the surveillance
of liver disorder using classification algorithms. Int J Comput Appl
2012;57(6):39–42.
[40] Santi Wulan Purnami, Mohamad Zain J, Embong Abdullah. A new ex-
pert system for diabetes disease diagnosis using modified spline smooth
support vector machine. In: ICCSA 2010; 2010.
[41] Santhanam T, Padmavathi MS. Application of K-means and genetic algo-
rithms for dimension reduction by integrating SVM for diabetes diagnosis.
Proc Comput Sci 2015;47:76–83.
[42] Vispute Nilesh Jagdish, Kumar Sahu Dinesh, Rajput Anil. An empirical
comparison by data mining classification techniques for diabetes data set.
Int J Comput Appl 2015;131(2):6–11.
[43] Peter S. An analytical study on early diagnosis and classification of dia-
betes mellitus. Bonfring Int J Data Min 2014;4(2):7–11.
[44] Velide Phani Kumar, Velide Lakshmi. A data mining approach for
prediction and treatment of diabetes disease. Int J Sci Invent Today
2014;3(3):73–9.
[45] Obaloluwa Olaniyi Ebenezer, Adnan Khashman. Onset diabetes diagnosis
using artificial neural network. Int J Sci Eng Res 2014;5(10):754–9.
[46] Aljumah Abdhullah A, Gulam Ahamad Mohammed, et al. Applications of
data mining: diabetes health care in young and old patients. J King Saud
Univ, Comput Inf Sci 2013;25:127–36.
[47] Marinov Miroslav, Mosa Abu Saleh Mohammad, Yoo Illhoi,
Boren Suzanne Austin. Data mining technologies for diabetes: a system-
atic review. J Diabetes Sci Technol 2011;5(6):1549–56.
[48] Nagarajan S, Chandrasekaran RM, Ramasubramanian P. Data mining
techniques for performance evaluation of diagnosis in gestational diabetes.
Int J Curr Res Acad Rev 2014;2(10):91–8.
[49] Lakshmi KR, Kumar SP. Utilization of data mining techniques for predic-
tion of diabetes disease survivability. Int J Sci Eng Res 2013;4(6):933–42.
323
[50] Masethe Hlaudi Daniel, Masethe Mosima Anna. Prediction of heart dis-
ease using classification algorithms. In: Proceeding of the world congress
on engineering and computer science; 2014.
[51] Dangare S Chaitrali, Apte Sulabha S. Improved study of heart disease pre-
diction system using data mining classification techniques. Int J Comput
Appl 2012;47(10):44–8.
[52] Sudha A, Gayathri P, et al. Effective analysis and predictive model
of stroke disease using classification methods. Int J Comput Appl
2012;43(14):26–31.
[53] Panzarasa Silvia, Quaglini Silvana, Sacchi Lucia, Cavallini Anna, Mi-
cieli Giuseppe, Stefanelli Mario. Data mining techniques for analyzing
stroke care processes. Stud Health Technol Inform 2010;160:939–43.
[54] Subbalakshmi G, Ramesh K, et al. Decision support in heart disease pre-
diction system using naïve bayes. Ind J Comput Sci Eng 2011;2(2):170–6.
[55] Krishnaiah V, Narsimha G, Chandra NS. Heart disease prediction system
using data mining technique by fuzzy KNN approach. Adv Intell Syst
Comput 2015:371–84.
[56] Bashir S, Qamar U, Hassan Khan F, Younus Javed M. MV5: a clinical dss
framework for heart diseases prediction using majority vote based classi-
fier ensemble. Arab J Sci Eng 2014;39:7771–83.
[57] Nishara Banu MA, Gomathy B. Disease forecasting system using data
mining methods. In: 2014 IEEE international conference on intelligent
computing applications; 2014. p. 130–3.
[58] Shouman Mai, Turner Tim, Stocker Rob. Applying k-nearest neighbour in
diagnosing heart disease patients. Int J Inf Educ Technol 2012;2(3):220–3.
[59] Chen Ling, Li Xue, Yang Yi, Kurniawati Hanna, Sheng Quan Z, Hu Hsiao-
Yun, et al. Personal health indexing based on medical examinations: a data
mining approach. Decis Support Syst 2014;81:54–65.
[60] Gupta Swati, Karandikar AM. Diagnosis of diabetic retinopathy using ma-
chine learning. J Res Dev 2015;3(2):1–6.
[61] Silva F Lincoln, Sequeiros Giomar O, Santos Maria Lúcia O,
Fontes Cristina AP, Muchaluat-Saade Débora C, Conci Aura. Thermal sig-
nal analysis for breast cancer risk verification. In: MEDINFO; 2015.
[62] Vaghela Chaitali, Bhatt Nikita, Mistry Darshana. A survey on various clas-
sification techniques for clinical decision support system. Int J Comput
Appl 2015;116(23):14–7.
[63] Ramachandran P, Girija N, Bhuvaneswari T. Early detection and pre-
vention of cancer using data mining techniques. Int J Comput Appl
2014;97(13):48–53.
[64] Evirgen Hayrettin, Çerkezi Menduh. Prediction and diagnosis of dia-
betic retinopathy using data mining technique. Turk Online J Sci Technol
2014;4(3):32–42.
[65] Kalairasai A, Mohammad Amanulla K. Unconscious oral cancer detection
using data mining classification approaches. Int J Adv Res Comput Eng
Technol 2015;4(7):3177–84.
[66] Kumar Mohanty Ashwini, Ranjan Senapati Manas, Kumar Lenka Saroj.
An improved data mining technique for classification and detec-
tion of breast cancer from mammographs. Neural Comput Appl
2013;22(1):303–10.
[67] Khraya Shweta. Using data mining techniques for diagnosis and prognosis
of cancer disease. Int J Comput Sci Eng Inf Technol 2012;2(2):56–66.
[68] Quellec Gwenole, Lamard Mathieu, Erginay Ali, Chabouis Agnes, Beat-
rice Massin Pascale, Cochener Azuguel Guy. Automatic detection of re-
ferral patients due to retinal pathologies through data mining. Med Image
Anal 2016;29:47–64.
[69] Chaudhary Kavita, Bajaj Pinki. Automated prediction of RCT (root canal
treatment) using data mining techniques: ICT in health care. In: Elsevier
proceeding of international conference on information and communication
technologies; 2014.
[70] Nandy Mahua. An analytical study of supervised and unsupervised clas-
sification methods for breast cancer diagnosis. In: 2nd international con-
ference on computing communication and sensor network. Int J Comput
Appl; 2013. p. 1–4.
[71] Razavi Amir R, Gill Hans, Åhlfeldt Hans, et al. A data pre-processing
method to increase efficiency and accuracy in data mining. In: AIME
2005. LNAI, vol. 3581. 2005. p. 434–43.
[72] Asgarnezhad Razieh, Shekofteh Maryam, Boroujeni Farsadzamani. Im-
proving diagnosis of diabetes mellitus using combination of preprocessing
techniques. J Theor Appl Inf Technol 2017;95(13).
324 M. Sharma et al. / IRBM 38 (2017) 305–324
[73] Ashok Kumar D, Govindsamy R. Performance and evaluation of classifi-
cation data mining techniques in diabetes. Int J Comput Sci Inf Technol
2015;6(2):1312–9.
[74] Jayalskshmi T, Santhakumaran A. Impact of preprocessing for diagnosis
of diabetes mellitus using artificial neural networks. In: Second interna-
tional conference on machine learning and computing; 2010. p. 109–12.
[75] Hamidi H, Daraei A. Analysis of preprocessing and post-processing
methods and using data mining to diagnose heart disease. Int J Eng
2016;29(7):921–30.
[76] Yan Hongmei, Zheng Jun, Jiang Yingtao, et al. Selecting critical clini-
cal features for heart diseases diagnosis with a real-coded GA. Appl Soft
Comput 2008;8:1105–11.
[77] Vembandasamy K, Karthikeyan T. Novel outlier detection in dia-
betes classification using data mining techniques. Int J Appl Eng Res
2016;11(2):1400–3.
[78] Heydari Mahmoud, Teimouri Mehdi, Heshmati Zainabolhoda, et al. Com-
parison of various classification algorithms in the diagnosis of type 2 dia-
betes in Iran. International J Diabetes Dev Countries 2016;36(2):167–73.
[79] Cibin NV, Franklin S Wilfred, Ajin NV. Diagnosis of diabetes mellitus and
NPDR in diabetic patient from tongue images using LCA classifier. Int J
Adv Res Trends Eng Technol 2015;2:57–62.
[80] Klein Barbara EK, Klein Ronald. Lifestyle exposures and eye diseases in
adults. Am J Ophthalmol 2007;144(6):961–9.
[81] http://www.youroralhealth.ca/oral-health-a-your-body/lifestyle.
[82] Sharma P, Busby M, Chapple RL, et al. The relationship between gen-
eral health and lifestyle factors and oral health outcomes. Br Dent J
2016;221(2):65–9.
[83] Kim Hyeongsoo, Ishag Musa Ibrahim M, Piao Minghao, et al. A data
mining approach for cardiovascular disease diagnosis using heart rate vari-
ability and images of carotid arteries. Symmetry 2016;8(6):1–15.