99% Accuracy Is Not Always Good: What Fraud Detection Taught Me About ML

My fraud detection model hit 99% accuracy and still failed. Here’s what imbalanced data, precision, recall, F1-score, confusion matrices, and SMOTE taught me the hard way.