Logistic Regression

Logistic Regression Q&A

1What is logistic regression?
Answer: Classification model estimating class probabilities via sigmoid function.
2Why not linear regression for classification?
Answer: Linear outputs are unbounded and not valid probabilities.
3What is sigmoid?
Answer: Maps real numbers to probability range (0,1).
4What are odds and log-odds?
Answer: Odds = p/(1-p); logistic model is linear in log-odds.
5Loss function?
Answer: Binary cross-entropy (log loss).
6How choose threshold?
Answer: Based on business cost trade-off using precision/recall curves.
7How interpret coefficients?
Answer: Change in log-odds per unit increase of feature.
8Role of regularization?
Answer: Prevents overfitting and stabilizes coefficients.
9Multiclass approach?
Answer: One-vs-rest or multinomial logistic regression.
10Common evaluation metrics?
Answer: Precision, recall, F1, ROC-AUC, PR-AUC.
11When logistic regression is strong?
Answer: With mostly linear decision boundary and interpretable requirements.
12One-line summary?
Answer: Logistic regression is a robust, interpretable baseline for classification.