Developing a Neural Network Model for Type 2 Diabetes Detection

Abstract

Worldwide, the healthcare system is greatly impacted by the changing requirements of the people. Diabetes is a long-lasting condition that can lead to serious complications if not controlled correctly. It is divided into Type 1 (DT1) and Type 2 (DT2) diabetes. Research shows that almost 90% of Diabetes cases are DT2, with DT1 making up around 10% of all Diabetes cases. This paper suggests a Rough-Neuro classification model for identifying Type 2 Diabetes, which includes a two-stage process. The approach includes utilising Rough sets JohnsonReducer to eliminate unnecessary features or characteristics and multi-layer perceptron for illness categorization. The suggested technique seeks to reduce the amount of input characteristics, which results in a reduction in the time needed to train the neural network and the storage space required. The findings show that decreasing the amount of input characteristics results in a lower neural network training time, enhances model performance, and reduces storage needs by 63%. It is worth mentioning that a smaller neural network with only seven hidden layers, trained for 1000 epochs with a learning rate of 0.01, attained the best performance, but time and storage were much decreased.