Logistic Regression
A linear model for classification that maps a linear score through a sigmoid to a probability.
- Despite the name, it is a classifier, not a regressor.
- Trained by minimizing log loss (cross-entropy).
- Coefficients are interpretable as log-odds ratios.
When to use: Default baseline for binary classification; calibrated probabilities and fast inference.
Example: Credit default: P(default) = σ(−2.5 + 0.04 · debt_ratio − 0.001 · income + 0.3 · late_payments). A threshold of 0.5 splits approve/decline.