AI-Driven Neural Networks for Early-Stage Diabetes Prediction

Abstract

Glucose intolerance is a common catabolism ailment that can lead to serious consequences such as cardiovascular disease, renal failure, and blindness. Nearly 77 million people in India have type 2 diabetes, and another 25 million are at risk of getting it. India has the second-highest rate of diabetes in the world. A lot of people still don't know about the health risks they face, which shows how important early detection is to lower death rates and improve patient health. The proposed methods deal with how well CNN, LSTM, and SimpleRNN models can forecast the early phase of diabetes. For this research, we collected live (primary source) data and preprocessed, we are making it as a standard dataset and we will publish it, tentatively named “Southern India Diabetes Dataset (SIDD)”. Our live dataset comprises 806 patient samples, of which 532 have diabetes patient samples and 274 are nondiabetes patient samples. We primarily considered demographic and clinical factors, such as age, gender, and diabetes symptoms. Various neural network models have trained the dataset. The models have achieved accuracy of 96% for the CNN, 95% for the LSTM, and 99.99% for the SimpleRNN. We evaluated the algorithms' performance based on the F1-score, recall, and precision. This indicates that modern deep learning models are proficient at distinguishing between individuals with diabetes and those who are not diabetic.