Evaluation Metrics for Classification and Regression

Evaluation metrics gauge model performance depending on whether your target is continuous (regression) or categorical (classification). Regression metrics measure the magnitude of prediction errors, while classification metrics measure the correctness of category assignments and the trade-offs between different error types.

Regression Metrics

These measure prediction errors for continuous values:

• MAE (Mean Absolute Error): Average absolute error; robust to outliers.
• MSE (Mean Squared Error): Average squared error; heavily penalizes outliers.
• RMSE (Root Mean Squared Error): Square root of MSE; interprets in target units while penalizing large errors.
• R² (R-Squared): Measures variance explained by the model, typically ranging from 0 to 1.

Classification Metrics

These evaluate predictions of distinct, categorical labels:

• Accuracy: Overall correct predictions; best for balanced classes.
• Precision: Focuses on correctness of positive predictions; crucial when false positives are costly.
• Recall (Sensitivity): Focuses on finding all actual positives; critical when false negatives are dangerous.
• F1-Score: Harmonic mean of precision and recall for imbalanced data.
• AUC-ROC: Measures discrimination capability across thresholds.