HGHH: A Feature-Driven Hybrid Gradient Boosting and Metaheuristic Optimization Framework for Predictive Intelligence in Industry 4.0

Abstract

In the context of Industry 4.0, data-driven predictive intelligence is vital for enhancing operational efficiency, early fault detection, and condition-based maintenance in smart manufacturing systems. However, conventional machine learning models often struggle with suboptimal feature selection and lack adaptability to dynamic production environments. This study proposes a novel hybrid framework HGHH that integrates Histogram Gradient Boosting Classification (HGBC) and Light Gradient Boosting Classification (LGBC) with two advanced metaheuristic optimizers: Horse Herd Optimization (HHO) and the Slime Mold Algorithm (SMA) for intelligent hyperparameter tuning. Feature dimensionality is reduced using Fast Correlation-Based Filter (FAST) and Class Activation Mapping (CAM) to retain critical predictive signals while improving model interpretability. Applied to a real-world smart manufacturing dataset, the HGHH framework achieved an accuracy of 0.987, precision of 0.981, recall of 0.993, and F1-score of 0.991, demonstrating its effectiveness in early anomaly detection and real-time decisionmaking. The model prioritizes high-impact features such as temperature variations and error metrics, ensuring robust performance, scalability, and resilience across diverse industrial settings