Majority Voting: Hybrid Classification Model for Diabetes Prediction

Diabetes, a chronic metabolic condition marked by abnormal blood glucose levels due to either inefficient insulin utilization or insufficient production, has spurred numerous academic efforts to develop reliable prediction models using machine learning (ML) algorithms. However, skepticism persists regarding ML applications in certain healthcare domains, mainly due to their opaque decision-making processes, which hinder widespread adoption by end-users. As a result, physicians often find it difficult to understand and trust these models for clinical use. This study proposes a majority voting-based hybrid modeling approach that employs Multilayer Perceptron (MLP), Random Forest (RF), and ExtraTrees (ExTr). Using the widely recognized PIMA Indians Diabetes dataset from the National Institute of Diabetes and Digestive and Kidney Diseases, the performance of each algorithm is assessed through metrics such as accuracy, precision, recall, and F-score. Additionally, the proposed hybrid voting model, which combines all preceding classifiers and determines predictions through a voting mechanism, is compared against the performance of individual classifiers. The results show that the proposed hybrid voting model demonstrates superior accuracy compared to individual classifiers. In terms of precision, recall, and F-score, the proposed model also outperforms the other classifiers.