Log Loss [Binary Classification]

Description

Use case: Classification problems (Binary classification)
When to use: Is specifically used for binary classification problems where the output is a probability (e.g., logistic regression). It quantifies the difference between the predicted probability and the true label (0 or 1).
Key property: Encourages probabilistic models to predict values that are closer to the true label (e.g., probability of 0 or 1). Assigning a higher predicted probability to the wrong class results in higher penalties.
Example applications:
- Predicting whether an email is spam or not
- Determining if a customer will churn (leave)

نمودار loss function:

چون مقادیر این مدل حتما بین 0 تا 1 است از بخش های بزرگتر از 1 در دو تصویر بالا صرف نظر شد

Normal:

Regularized: