Logistic regression predicts a binary outcome with interpretable coefficients and well‑calibrated probabilities for decision thresholds.
Interpretation
Coefficients translate to odds ratios (exp(coeff)). Provide CIs to communicate uncertainty. Standardize features when comparing magnitudes.
Regularization
- L1 (Lasso) for sparsity; L2 (Ridge) for stability; elastic‑net mixes both
- Cross‑validate penalty strength and inspect coefficient paths
Metrics
- ROC‑AUC and PR‑AUC (for imbalance)
- Confusion matrix at chosen threshold, F1/precision/recall
- Calibration curve and Brier score
Class Imbalance
Use class weights or resampling (e.g., SMOTE) and prefer PR‑AUC over ROC‑AUC when positives are rare.