Predictive Maintenance: Recall-Optimized Failure Forecasting

A machine learning pipeline designed to detect rare equipment failures in high-imbalance manufacturing datasets.

Key Objectives:

• Minimize Downstream Cost: Prioritized Recall over Accuracy to penalize false negatives (missed failures) more than false positives (unnecessary checks).
• Interpretability: Utilized SHAP values to engineer features like Wear_x_Torque and isolate collinearity between air/process temperatures.
• Model Stacking: Implemented a Voting Classifier (Random Forest + XGBoost) to achieve higher Recall.