Comparative Study of Machine Learning Approaches Based on Artificial Neural Network, Regression, and Clustering for Diabetes Prediction

Abstract

This study presents a comparative analysis of three machine learning model and algorithms Artificial Neural Network (ANN), Logistic Regression, and K-Means Clustering using the Pima Indians Diabetes dataset. The main objective is to evaluate the performance of supervised and unsupervised methods in predicting diabetes based on physiological and clinical features. he ANN model was developed using a feedforward and backpropagation approach, Logistic Regression applied the fundamental logit equation, and K-Means Clustering was employed as an unsupervised reference. Model performance was assessed using Accuracy, Precision, Recall, and F1-score for supervised models, and Adjusted Rand Index (ARI) for clustering. Experimental results indicate that Logistic Regression achieved the best accuracy of 0.7573, followed by ANN with 0.7078, while K-Means obtained an ARI of 0.1614. The heatmap comparison shows that supervised models outperform unsupervised approaches, with Logistic Regression offering better interpretability and stability, and ANN demonstrating the ability to model nonlinear relationships. K-Means, though less accurate, provided valuable insight into data structure and natural grouping. Overall, the findings confirm that supervised learning models, particularly Logistic Regression and ANN, are more effective for medical prediction tasks. Future research may explore hybrid or ensemble models that combine the interpretability of Logistic Regression, the adaptability of ANN, and the exploratory capability of clustering to enhance medical diagnostic performance.