Deep Learning based System Design for Diabetes Prediction

Abstract

Amongst various chronic ailments, in recent years Diabetes became a diseases which is major cause of fatalities in today’s world. Therefore, a timely prediction of the symptoms of the diabetes significantly plays a great role in reduction of the mortality rate due to diabetes. Nowadays, a large amount of diabetes data is available in different data repositories such as the kaggle, MNIST and UCI. Diabetes is becoming majorly a world's most common, chronic, owing to complications, dreaded diseases. Diabetes must be detected early in order to receive appropriate treatment and to prevent the illness from progressing. Not only can the suggested approach to be utilized in forecasting incidence of diabetes in the future, however it can be applied to find out type of diabetes a person possesses. With various changes in treatment approaches between type 2 and type 1 diabetes, such schemes will aid in providing best therapy for patients. Our prototype is basically designed by applying concealed layers of a deep neural network and uses dropout regularization to avoid over fitting by converting job into a classification issue. We tuned a few parameters and utilised the binary cross-entropy loss function to create a high-accuracy deep neural network prediction model. The experimental findings demonstrate the efficacy and suitability of the proposed (Deep Learning for Diabetes Prediction) model. The Pima Indians diabetes data set has the best training accuracy of 98.07 percent. The Pima Indians diabetes and diabetic type databases have been subjected to extensive testing. The experimental findings demonstrate that our suggested model outperforms current techniques.