This study extends logistic regression model to fuzzy logistic regression model. It suggests a new approach based on fuzzy concepts to estimate model parameters. The proposed models outperform ML approach for small size datasets.
This study extends logistic regression model to fuzzy logistic regression model. It suggests a new approach based on fuzzy concepts to estimate model parameters. The proposed models outperform ML approach for small size datasets.
This study extends logistic regression model to fuzzy logistic regression model. It suggests a new approach based on fuzzy concepts to estimate model parameters. The proposed models outperform ML approach for small size datasets.
Parameter estimation for logistic regression is usually based on maximizing the
likelihood function. For large well-balanced datasets ML estimation is a satisfactory approach. Unfortunately, ML may fail completely or at least produce poor results in terms of estimated probabilities and confidence intervals of parameters, specially for small datasets. This study extends logistic regression model to fuzzy logistic regression model by suggesting a new approach based on fuzzy concepts to estimate the model parameters. This study produces three proposed mathematical models with different objective functions. The first is formulated as a bi-objective programming model. The second is formulated using a goal programming approach, while the third is a mathematical programming model which minimizes the total spread of the estimated probabilities of the logistic model. The proposed models are evaluated and their results are compared to ML results through a Monte Carlo simulation study. The results are analyzed and summarized to conclude the following: The proposed models outperform ML approach for small size datasets with respect to the similarity measure as goodness of fit index. Meanwhile, they produce a reasonable level of overall rate of correct classification. The proposed models perform successfully for small datasets. Moreover, they do not fail if the average value of Y is exactly 0 or 1. The performance of the Goal Programming Model for fuzzy logistic regression outperforms the Fuzzy Logistic Model with minimum fuzziness and the Bi-objective Model respectively.
Key Words
Logistic Regression, Maximum Likelihood, Small Data Sets,