[FAQ]Why should we use predict_proba instead of predict when computing ROC AUC? #477

Course

machine-learning-zoomcamp

Question

When evaluating a Logistic Regression model with ROC AUC, why should we use the probabilities from predict_proba() instead of the class labels from predict()?

Answer

ROC AUC evaluates how well the model ranks positive examples above negative examples across different thresholds.

predict() returns only the final class labels, such as 0 or 1. This removes most of the information about how confident the model is.

For ROC AUC, we should instead use the probability of the positive class:

y_pred = model.predict_proba(X_val)[:, 1]

auc = roc_auc_score(y_val, y_pred)

These probability scores allow ROC AUC to evaluate the ranking of predictions across different thresholds.

So, for ROC AUC, predict_proba()[:, 1] is more appropriate than predict().

Checklist

• I have searched existing FAQs and this question is not already answered
• The answer provides accurate, helpful information
• I have included any relevant code examples or links