A Hybrid QPSCO Metaheuristic for Joint Feature Selection and Hyperparameter Tuning in Early Stage Diabetes Prediction

Abstract

Type 2 Diabetes Mellitus (T2DM) continues to pose a serious health challenge. Although Machine Learning has been widely employed for disease prediction, its effectiveness depends heavily on feature selection (FS) and hyperparameter tuning (HPT). This study proposes Quantum Particle Swarm + Single Candidate Optimizer (QPSCO), a novel hybrid metaheuristic that combines the global search capability of Quantum Particle Swarm Optimization (QPSO) with the strong local exploitation ability of the Single Candidate Optimizer (SCO). The proposed approach jointly performs FS to reduce dimensionality and HPT to improve classifier performance. Experiments were conducted on the “Early Stage Diabetes Risk Prediction” dataset using several Machine Learning classifiers. The results show that QPSCO-optimized models consistently achieved competitive or superior performance in terms of accuracy, precision, F1-score, and AUC-ROC compared with related studies conducted under similar experimental conditions. QPSCO-XGBoost achieved up to 96.82% accuracy, 98.84% precision, 97.66% F1-score, and 98.32% AUC-ROC using nine selected features, while QPSCO MLP achieved 96.42% accuracy and 98.43% AUC-ROC using only eight selected features under traditional cross-validation. Comparative evaluation showed improved performance over existing methods. Stratified Nested Cross-Validation (SNCV) with SMOTE-NC provided more reliable performance estimates. By selecting the most informative features, the proposed approach proved effective for early T2DM detection.

Keywords