Diabetes is a chronic condition. It occurs when the body cannot use insulin effectively or produce enough of it. This leads to high blood sugar. If left undetected, it can cause severe complications. Thus, advanced predictive systems are needed in e-Health. This study proposes an intelligent hybrid system that integrates machine learning clustering techniques (K-Means, Complete-Linkage, Expectation-Maximisation, and Hierarchical K-Means) with Ant Colony Optimisation (ACO) for feature selection, aiming to optimise early-stage diabetes prediction in next-generation computing paradigms. We use a preprocessed dataset. It includes key risk indicators like polydipsia and polyuria. The system removes irrelevant features. This improves computational efficiency. As a result, it becomes suitable for IoT-enabled smart health applications and cyber–physical systems. Models are evaluated using precision, recall, Rand index, and Fowlkes-Mallows score, revealing significant improvements: Expectation-Maximisation achieves an 81.45% recall increase and 97% precision with ACO, while Hierarchical K-Means improves recall by 64.93%. False negatives drop notably (e.g., −101 for Expectation-Maximisation), demonstrating reduced computational overhead and higher accuracy. This approach not only advances e-Health through reliable clinical decision-making but also contributes to sustainable computing by enabling efficient AI deployment in edge-based or cloud-integrated health monitoring systems, with potential extensions to online tools and broader medical datasets. • Hybrid clustering + ACO feature selection model achieves 97% precision in early diabetes prediction. • ACO feature selection cuts false negatives (−101 for EM) and boosts clustering performance. • ACO improves accuracy of K-Means, EM Complete-Linkage recall drops 38.24%. • Key diabetes features: Polydipsia, Polyuria; suggests online tools & wider medical dataset use.
Khan et al. (Sun,) studied this question.