A Cyber-Physical Analytical Model for Industrial Machine Fault Prediction Using a Hybrid AI Technique

Abstract

Predictive maintenance can aid in early failure detection and minimize unplanned industrial downtimes. In this study, the authors build a machine-failure prediction pipeline based on Random Forest (RF) in the AI4I 2020 Predictive Maintenance dataset. This framework integrates leakage-aware preprocessing, feature selection using importance, balanced class weighting, and hyperparameter optimization. It was compared to the support vector machine with the standard held-out test protocol. On all of the reported metrics, the integrated RF pipeline's accuracy measured 92%, precision 90%, recall 89%, and F1-score 90%, which is superior to the baseline. The confusion-matrix analysis also revealed reduced false alarms and failures. The results show the feasibility of the integrated RF pipeline as a predictive-maintenance decision support approach. Before deployment, further validation with repeated cross-validation, further benchmark models, and real industrial data is needed.