Prognosis of diabetes using fuzzy inference system and multilayer perceptron

Abstract

Nowadays, due to busy lifestyle of people, ignorance and tolerance of symptoms, slowly develops diabetes. It cause when body stops producing enough insulin which is essential to control the blood sugar. The amount of high sugar level affects vital organs of the body like kidney, heart, and brain. The diagnosis of diabetes typically confirmed on the basis of increased glucose level in the blood. Symptoms and risk factors also play a major role for diagnosis. Such kind of medical diagnosis problem can be solved by combining the fuzzy systems and neural network. This paper presents the novel approach for prognosis of type-2 diabetes & prediabetes using FIS and MLP. The FIS used here for predicting the initial risk of prediabetes and type-2 diabetes using blood tests, to measure the sugar/glucose levels in different situations like fasting, post meal and random glucose. The output of the FIS, related symptoms, and risk factors are used to train the perceptron network, which results in one of the class as non-diabetes, prediabetes and type-2 diabetes. The proposed model is trained and tested on 385 patient's data and gives 91% accuracy of classification, specificity about 94% and sensitivity 91%. This proposed system will helps the medical practitioner for diagnosis of diabetes.