Predictive Modeling of Diabetes Classification using Artificial Neural Networks

Abstract

Diabetes mellitus is a chronic disease that has become a major global public health challenge. Timely and accurate prediction not only aids in immediate treatment but also plays a crucial role in formulating effective strategies. In this research, we compare the prediction of diabetes using two popular neural network models Multi-Layer Perceptron (MLP) and General Regression Neural Network (GRNN). This analysis is based on the PIMA Indian Diabetes Dataset, which contains medical information of female patients, including glucose levels, BMI, insulin amount, age, etc. In this research, the dataset underwent stages of cleaning, normalization, and division into training and testing sets. Then, the mathematical details of the MLP and GRNN models were described, which included forward propagation, activation functions, and loss formulas. The MLP model used two hidden layers with ReLU and Sigmoid activation functions, while the GRNN model used Gaussian radial basis functions and Euclidean distance. After training, the performance of both models was evaluated using metrics such as accuracy, confusion matrix, and ROC-AUC, revealing that both models proved effective in predicting diabetes. However, GRNN demonstrated better overall performance due to its non-repetitive structure and smooth results, while MLP exhibited more effective adaptation thanks to fast computation and deep network architecture, making it suitable for large and complex data. This research indicates that if ANN models are designed and configured correctly, they can help in the timely and effective diagnosis of diabetes. This analysis provides guidance to healthcare professionals and data scientists in selecting appropriate models.