Performance Evaluation of Different Artificial Neural Network Models in the Classification of Type 2 Diabetes Mellitus

Authors : Emek GÜLDOĞAN, Zeynep TUNÇ, Ayça ACET, Cemil ÇOLAK

Pages : 23-32

View : 16 | Download : 7

Publication Date : 2020-06-30

Article Type : Research Paper

Abstract :Objective: In this study, it is aimed to classify type 2 Diabetes Mellitus insert ignore into journalissuearticles values(DM);, compare the estimates of the Artificial Neural Network models and determine the factors related to the disease by applying Multilayer Perceptron insert ignore into journalissuearticles values(MLP); and Radial Based Function insert ignore into journalissuearticles values(RBF); methods on the open-access dataset. Material and Methods: In this study, the data set named “Pima Indians Diabetes Database” was obtained from https://www.kaggle.com/uciml/pima-indians-diabetes-database. The dataset contains 768 records with 268 insert ignore into journalissuearticles values(34.9%); type 2 diabetes patients and 500 insert ignore into journalissuearticles values(65.1%); people without diabetes, which have 9 variables insert ignore into journalissuearticles values(8 inputs and 1 outcome);. MLP and RBF methods, which are artificial neural network models, were used to classify type 2 DM. Factors associated with type 2 DM were estimated by using artificial neural network models. Results: The performance values obtained with MLP from the applied models were accuracy 78.1%, specificity 81.2%, AUC 0.848, sensitivity 71%, positive predictive value 61.7%, negative predictive value 86.8% and F-score 66%. In relation to RBF model, the performance metrics were accuracy obtained 76.8%, specificity 82.1%, AUC 0.813, sensitivity 66.0%, positive predictive value 64.6%, negative predictive value 83% and F-score 65.3%, respectively. When the effects of the variables in the data set examined in this study on Type 2 DM are analyzed; The three most important variables for the MLP model were obtained as Glucose, BMI, Pregnancies respectively. For RBF, it was obtained as Glucose, Skin Thickness, and Insulin. Conclusion: The findings obtained from this study showed that the models used gave successful predictions for Type 2 DM classification. Besides, unlike similar studies examining the same dataset, the significance values of the factors associated with the models created were estimated.
Keywords : Classification, Multilayer perceptron neural network, Radial based function neural network, Type 2 Diabetes Mellitus