Enhancing Diabetes Disease Prediction and Privacy Preservation via Federated Learning and PSO-WCO Optimization

Abstract

Diabetes mellitus is a leading non-communicable disease, affecting over 537 million individuals globally. Its progression, often influenced by obesity and genetic factors, poses significant health risks, including cardiovascular, renal, and neurological complications. Early detection is essential to minimize these risks. This study addresses class imbalance using Synthetic Minority Over-sampling Technique (SMOTE) and evaluates various classifiers, with AdaBoost achieving the best performance (94.02\% accuracy, 93.32\% F1 score, and 0.95 AUC). To further enhance prediction while preserving data privacy, a novel Federated Learning with Particle Swarm Optimization (FLPSO) model is introduced. In centralized learning, AdaBoost combined with PSO-WCO (Particle Swarm Optimization -Weighted Conglomeration Optimization) attained 96.40\% accuracy, while FLPSO in a federated setup achieved 98.30\%, surpassing existing methods. The proposed model effectively balances prediction accuracy, data privacy, and communication efficiency, highlighting its potential in secure and reliable diabetes prediction and its applicability to related health risk assessments.