Prediction of diabetes mellitus at early stage using deep learning methods

Abstract

Diabetes Mellitus is one of the world’s leading cause of heart attacks, kidney failure, lower limb ablation, stroke and blindness. The number of affected people increased from 108 million to 422 million from 1980 to 2014 and will go as high as 629 million by 2045.Automated testing, early prediction and diagnosis of diabetes mellitus is therefore essential for improvement of patient’s survival rate. In this paper, we have explored various deep learning methods like single LSTM, Auto-Encoder, optimized CNN + LSTM and LSTM Skip connection model over the same two datasets PIMA Indian Diabetes dataset and dataset collected from the victims in Sylhet Diabetes Hospital, Bangladesh. We have achieved highest accuracy of 93.26% in LSTM with Skip connection algorithm and highest Precision of 98.53% in Optimized LSTM + CNN algorithm.