Intelligent Failure Detection in Industrial Machines Using Hybrid ML and Anomaly Scores

Predictive maintenance (PdM) has become a game-changing strategy for improving industrial efficiency and reliabil-ity. The traditional time-based maintenance further increases the amount of money spent while not preventing failures. Unplanned machine failures in industrial settings can result in serious down-time, monetary losses, and safety risks. The efficacy of advanced machine learning is limited in manufacturing by the lack of labelled machine data and heterogeneous data sources, which hinder predictive maintenance. In order to address these issues, this study proposes a hybrid machine learning framework that effectively integrates supervised learning and unsupervised learning to enhance predictive maintenance. By combining anomaly detection with supervised classification, the model enables early prediction of machine failure, thus facilitating efficient data-based decision-making. Anomaly scores generated using the Isolation Forest algorithm help uncover hidden failure patterns and improve classification robustness. To improve the learning signal, these scores are added as extra features. The training set is subjected to the SMOTE. This balanced and enriched dataset is used to train a Random Forest classifier, Logistic Regression, and XGBoost. A baseline Random Forest model trained without anomaly scores is used to compare the suggested hybrid model. According to comparative analysis, the hybrid model achieves better performance than individual classifiers as well as the baseline model without anomaly integration. The proposed model achieved 98.15% accuracy and 0.988 ROC-AUC. This method thus leads to an effective and scalable solution for predictive maintenance in smart manufacturing environments.