Diabetes or Diabetes mellitus is a chronic lifelong medical condition that causes high sugar level in the blood. It is a life threatening disease that affects the human body’s ability to process glucose. Early and accurate diagnosis of diabetes is evident in other to reduce further degeneration of the condition. In this paper we used adaptive neuro fuzzy system to predict the diagnostic outcome of diabetes. Five different membership function (Triangular, Trapezoidal Generalized bell, Gauss1 and Guass2 membership functions) were used in the fuzzification layer of the Adaptive Neuro Fuzzy Inference System (ANFIS). The outcome of the experiment showed that Guass2 membership function generated the best result.
Keywords: ANFIS, Diabetes, Membership function
Diabetes or Diabetes mellitus is a chronic lifelong medical condition that causes high sugar level in the blood. It is a life threatening disease that affects the human body’s ability to process glucose. Early and accurate diagnosis of diabetes is evident in other to reduce further degeneration of the condition. In this paper we used adaptive neuro fuzzy system to predict the diagnostic outcome of diabetes. Five different membership function (Triangular, Trapezoidal Generalized bell, Gauss1 and Guass2 membership functions) were used in the fuzzification layer of the Adaptive Neuro Fuzzy Inference System (ANFIS). The outcome of the experiment showed that Guass2 membership function generated the best result.
Keywords: ANFIS, Diabetes, Membership function
Diabetes or Diabetes mellitus is a chronic lifelong medical condition that causes high sugar level in the blood. It is a life threatening disease that affects the human body’s ability to process glucose. Early and accurate diagnosis of diabetes is evident in other to reduce further degeneration of the condition. In this paper we used adaptive neuro fuzzy system to predict the diagnostic outcome of diabetes. Five different membership function (Triangular, Trapezoidal Generalized bell, Gauss1 and Guass2 membership functions) were used in the fuzzification layer of the Adaptive Neuro Fuzzy Inference System (ANFIS). The outcome of the experiment showed that Guass2 membership function generated the best result.
Keywords: ANFIS, Diabetes, Membership function
Benin Journal of Advances in Computer Science, www.bjacs.org Vol. 3. No. 1. Issue 3—December, 2018
MEMBERSHIP FUNCTION PERFORMANCE OF ADAPTIVE NEURO
FUZZY
SYSTEM ON DIABETES PREDICTION
Amadin, I. F.
Department of Computer Science
University of Bi
Edo State, Nigeria
E-mail:frankamadin@uniben.edu
Bello, M. E.
Department of Computer Science
University of Benin
ABSTRACT
Diabetes or Diabetes mellitus is a chronic lifelong medical condition that causes high sugar level in the
blood. It is a life threatening disease that affects the human body’s ability to process glucose. Early and ac-
curate diagnosis of diabetes is evident in other to reduce further degeneration of the condition. In this paper
we used adaptive neuro fuzzy system to prediet the diagnostic outcome of diabetes. Five different member-
ship fimetion (Triangular, Trapezoidal Generalized bell, Gauss! and Guass2 membership functions) were
used in the fuzzification layer of the Adaptive Neuro Fuzzy Inference System (ANFIS). The outcome of the
experiment showed that Guass2 membership function generated the best result,
Keywords: ANFIS, Diabetes, Membership function
1.0 INTRODUCTION
There are three type of diabetes; Type | Diabetes which is caused by the inability of the body to produce in-
sulin, Type 2 diabetes which arises from the inability of the body to produce sufficient insulin and gestational
diabetes which only affect women during pregnancy. In 2013 it was estimated that the number of persons
suffering from diabetes in the world is about 382 million and this number is on the rise (Rian et al, 2015)
Kaul predicted that the prevalence of diabetes will increase by 55% by the year 2035 (Kaul et al, 2015). Dia-
betes is known to affect individuals of all ages but its occurrence is mostly common amongst adult and the
older a person gets the more susceptive he/she becomes to diabetes. Symptoms of diabetes includes increased
thirst, insatiable appetite, dry mouth, frequent urination, unexplained weight loss, blurred vision, fatigue and
might include impotency. The pathogenesis of diabetes is still vague although studies have shown that com-
plicated ailments such as diabetes might be caused by life style and genetic factor (Rian et al, 2015 and
Shivakumar et al, 2014). Diabetes can be diagnosed medically using either of the fowling test, Hemoglobin
AIC test, Fasting Plasma Glucose (FPG) Test or the Oral Glucose Tolerance Test (OGTT), although various
computing techniques have been proposed for its diagnosis. Early and accurate diagnosis of diabetes is the
first step in curtailing the menace of the disease. “Adaptive Neuro Fuzzy Inference System (ANFIS) which
is a type of neuro fuzzy can play a vital role in medical diagnosis to curb the inaccuracies due to external hu-
man factors” (Appiah et al, 2015). ANFIS derived its term from Adaptive Network Fuzzy Inference System
and it was proposed by Jang in 1993, it was later retitled Adaptive Neural Fuzzy Inference System (Jyh-
Shing et al, 1993). “ANFIS refers to an inference system that integrates the best features of neural network
and fuzzy logic” (Piero, 2000). It is a system that predicts the relationship that exist in a set of data. “ANFIS
1Benin Journal of Advances in Computer Science, www.bjacs.org Vol. 3. No.1. Issue 3—December, 2018,
designed to allow IF-THEN rules and membership functions to be constructed based on the
historical data and also include the adaptive nature for automatic tuning of the membership
functions” (Jang et al, 1997). Studies have shown that the type of membership function utilized
by ANFIS determines its prediction accuracy (Talpur et al, 2017). In this paper an Adaptive neuro fuzzy infer-
ence system was used in diagnosing diabetes using variety of membership functions with an aim of improving
the accuracy of diabetes prediction.
2.0 REVIEW OF RELATED WORKS
In this section we discussed some related research on the use of artificial intelligence for diagnosing diabetes.
In 2010 Jayalakshmi et al used artificial neural network (ANN) to increase the diagnostic classification accura-
cy of diabetes (Jayalakshmi et al, 2010). This study was further improved by Thirugnanam et al, in 2012, by
combining ANN and fuzzy logic, The neuro fuzzy system was able to classify 89% of the test case accurately,
With the combination of both techniques it will be capable of learning and interpreting data but the diagnostic
accuracy is quite low.
In 2013, Meng et al conducted a study to compare the performance of three machine learning technique in di-
agnosing diabetes. They compared decision tree, artificial neural network and logistic regression using 12
symptom criteria, ‘The finding of their study revealed that the decision tree had the highest prediction accura-
cy. The decision tree in incapable of generating diagnosis for new problems and can only predict based on the
outcomes already created in the decision tree.
In 2011, Lee et al, designed an expert system using the fuzzy logic technique in diagnosing diabetes. In their
study the expert system was developed using five-layer fuxzzy ontology and this was used to model the system
under uncertainty and it extended the ontology domain for the system. Despite the suecess achieved by the sys-
tem in it incapable of learning from the patient data.
Other nature inspired techniques have also been applied in predicting diabetes. In 2012 Aribarg et al, combine
artificial bee colony algorithm and fuxzzy logic in predicting diabetes. In their study the artificial bee colony
algorithm was used to extract the best rules for the fuzzy system. Despite the success achieved by the system
in it incapable of learning from the patient data.
In a similar study conducted by Ganji et al, in 2011, they combined fuzzy logic and ant colony optimization
algorithm, The ant colony classification algorithm was used to extract the best rules of fuzzy and this made the
prediction of diabetes more accurate. Despite the success achieved by the system in it incapable of learning
from the patient data.
3.0 METHODOLOGYBenin Journal of Advances in Computer Science, www.bjacs.org Vol. 3. No. 1. Issue 3—December, 2018
3.1 ADAPTIVE NEURO FUZZY INFERENCE SYSTEM
Adaptive Neuro-Fuzzy Inference System (ANFIS) is the combination Artificial Neural Net-
work (ANN) and Fuzzy Logic (FL). It is a system that predicts the relationship that exist in a set of data. AN-
FIS has six layer which are explained below;
Layer 1: it allows input into the ANFIS
Layer 2: it computes the membership function of the variable. An example of bell membership function is giv-
en below
By) = @
Layer 3: it computes the fuzzy rules using antecedent values
uw.) * lv) *
u(vn) 2)
Layer 4: normalizes the fuzzy membership function
o
of =. __
"OF +03 + +++ 03
(3)
Layer 5: It converts the fuzzy data into actual output
Of = 070, (¥,) + Pa(v2) + Pan) +7) a
Layer 6; It generates the output of the system by summing the defuzzified output
of = » of (5)
3.2 MEMBERSHIP FUNCTIONS
‘A fuzzy set can only be described by a membership function. The value range for membership function is be-
tween 0 and 1. “A fuzzy set can therefore, unlike a crisp one, be represented by an infinite number of member-
ship functions” (Omar et al, 2015). “The simplest membership functions are formed using straight
lines” (Mandani, 1974). ‘There are various kinds of membership function, they include Triangular membership
function, Trapezoidal membership function, Generalized bell membership function, Gaussian membership
function, S-Shaped membership Function and x- Shaped membership Function, However the triangular and
trapezoidal membership functions are mostly used because of their simplicity.
3.3 MATERIALS
The dataset used in this study was gotten from https://archive.ics.uci.edu/ml/datasets/diabete, 100 cases wer
used in the simulation with 60% of the dataset was used to train the system while 40% was used to test the sys=
tem, Five different membership functions (Triangular, Trapezoidal Generalized bell, Gauss! and Guass2 mem-
3