SECOM Datset

Built a classification pipeline on the SECOM industrial dataset (1,567 samples, 590 sensor features) to predict manufacturing defects under severe class imbalance (6.6% failure rate). Compared Logistic Regression, Random Forest, and SVM using imbalance-aware metrics (Precision/Recall/F1/ROC-AUC), applying SMOTE and threshold tuning to raise defect-detection recall to 76%.