What is Log Loss?

Updated May 17, 2026

Short answer

Log loss measures the performance of classification models based on predicted probabilities.

Deep explanation

Lower log loss indicates better probability calibration. It penalizes incorrect confident predictions heavily.

Real-world example

Used in Kaggle competitions for evaluating classifiers.

Common mistakes

  • Using accuracy alone without probability evaluation.

Follow-up questions

  • Why is log loss differentiable?
  • Can log loss be negative?

More Logistic Regression interview questions

View all →