Regression Metrics

Regression Evaluation

Regression metrics quantify prediction error in different ways.

Mean Absolute Error

MAE = mean(|y_true - y_pred|). Robust to outliers. In same units as target. mean_absolute_error(y_true, y_pred).

Mean Squared Error

MSE = mean((y_true - y_pred)²). Penalizes large errors heavily. In squared units. mean_squared_error(y_true, y_pred).

RMSE = sqrt(MSE). Back in original units.

R-Squared

R² measures explained variance. 1.0 is perfect, 0.0 means predictions equal mean. r2_score(y_true, y_pred).

Adjusted R² accounts for number of predictors. Prevents inflation from adding features.

Key Takeaways

1. MAE is robust to outliers; MSE penalizes large errors
2. R² shows proportion of variance explained
3. Multiple metrics provide complete picture