Machine Learning-Based Predictive Maintenance in Industry: Fault Classification System

IMPACT is a machine learning-based predictive maintenance decision-support system developed for Roketsan A.Ş. to detect early signs of machine degradation and trigger timely maintenance alerts in safety-critical industrial environments. Seven supervised classification algorithms including SVM, Random Forest, and Gradient Boosting were trained on turbofan engine datasets. A key contribution is a timing-aware evaluation framework and a three-stage alert policy combining moving-average smoothing, probability thresholding, and persistence filtering. The system achieves accuracy exceeding 0.97 and F1-scores above 0.83.