Hybrid Particle Swarm Optimization and Genetic Algorithm for Diabetes Prediction

Abstract

Diabetes is a prevalent chronic illness across the world, and early detection has proven to be of great importance to enhance patient well-being. Machine learning algorithms have shown strong performance in forecasting diabetes, especially MLP neural networks. However, MLP prediction accuracy depends heavily on selecting optimal hyperparameters, which often presents an optimization challenge. To enhance the performance of MLPs in predicting diabetes, this study proposes an enhanced adaptive hybrid framework combining particle swarm optimization and a genetic algorithm (AHPSO-GA). The particle swarm optimization (PSO) algorithm is used to efficiently explore the search space and identify the best hyperparameter configurations for the MLP model. The genetic algorithm (GA) is applied to overcome the early convergence and diversity limitations associated with the PSO algorithm, periodically applying the best candidate solutions generated by the PSO algorithm. Furthermore, an adaptive strategy is introduced to dynamically adjust the mutation rate as the optimization process progresses. Specifically, when accuracy does not improve for a given number of iterations, the mutation rate is increased to encourage exploration; otherwise, it is decreased to promote the exploitation of high-quality solutions. The proposed AHPSO-GA-MLP framework was evaluated using a diabetes-specific dataset and compared to traditional MLP models. The experimental results showed that the proposed method achieved performance for accuracy was 84.0%, Precision 81.48%, Recall 88.0%, area under the curve (AUC) 87.60%, and F1-Score 84.62%.