An Enhanced Ant Colony Optimization Approach for Multi-Classifier Predictive Maintenance in Industrial IoT Systems

The Industrial Internet of Things (IIoT) enables continuous real-time monitoring of equipment and creates opportunities for data-driven predictive maintenance, but high-dimensional sensor streams and noisy signals make reliable failure prediction difficult. This paper presents a hybrid predictive-maintenance framework that combines an Enhanced Ant Colony Optimization (ACO) algorithm for joint feature selection and hyperparameter tuning with three supervised classifiers: Decision Tree (DT), Support Vector Machine (SVM), and Logistic Regression (LR). The enhanced ACO uses bounded pheromone limits and randomized search over parameter ranges to improve convergence and avoid feature dominance while identifying compact, informative feature subsets from IIoT sensor data (e.g., temperature, rotational speed, torque, tool wear). Optimized feature sets and classifier parameters are then used for binary failure prediction. We evaluate the framework on the AI4I 2020 Predictive Maintenance dataset: ACO-optimized models substantially improve performance over baselines, with ACO and DT achieving 99.91% accuracy, 99.92% precision, and 99.90% recall; ACO with SVM and ACO with LR perform comparably. Results indicate that swarm-based joint optimization paired with diverse classifiers yields a scalable, interpretable, and highly effective approach for IIoT predictive maintenance.