Choosing the Right Metric

Decision Tree: Which Metric to Use?

├─ Problem Type?
│  ├─ Regression (Predicting continuous values)
│  │  ├─ RMSE: Interpretable, penalizes outliers
│  │  └─ MAE: Robust to outliers, interpretable
│  └─ Classification (Predicting categories)
│     ├─ Balanced dataset?
│     │  ├─ Yes: Accuracy, Precision, Recall, F1, AUC-ROC
│     │  └─ No (Imbalanced):
│     │     ├─ Lightly imbalanced: AUC-ROC
│     │     └─ Heavily imbalanced: Precision-Recall AUC
│     └─ Cost of mistakes?
│        ├─ False Positives worse (Spam filter): Precision high
│        ├─ False Negatives worse (Cancer detection): Recall high
│        └─ Both equal: F1-Score