Course contentsShow
← Computer Science
Lesson 2837 of 2,87263. Classical ML AlgorithmsFree lesson

Assumptions and Limitations of Logistic Regression

Recognize when logistic regression is appropriate and understand its linearity and independence assumptions.

Assumptions and Limitations of Logistic Regression

What you'll learn: When logistic regression is the right tool for the job and what constraints it operates under.

Core Assumptions

Logistic regression makes several important assumptions that must hold for the model to work well:

Linear relationship in log-odds space: While logistic regression handles non-linear probability curves (thanks to the sigmoid function), it assumes the log-odds (logit) of the outcome have a linear relationship with features. If p is the probability of class 1, then log(p/(1-p)) = β₀ + β₁x₁ + β₂x₂ + ... must be approximately linear.

Feature independence: Logistic regression assumes features aren't highly correlated with each other (no multicollinearity). When features are redundant, coefficients become unstable and interpretation suffers—imagine trying to credit separate contributions when two features always move together.

Sufficient sample size: You need enough examples relative to your number of features. A rough rule: at least 10-15 observations per feature, and adequate representation of both classes.

Key Limitations

Linear decision boundaries: Without feature engineering (polynomial features, interactions), logistic regression can only draw straight lines (or hyperplanes) between classes. It can't naturally capture complex, curved boundaries.

Sensitivity to outliers: Extreme values can distort the fitted boundary, especially without regularization.

Binary or categorical outcomes only: Logistic regression predicts discrete classes, not continuous values—that's linear regression's domain.

When to Use Logistic Regression

Choose logistic regression when you need interpretable coefficients, have approximately linearly separable classes in feature space, and want probabilistic predictions with well-calibrated confidence.

Key Takeaway: Logistic regression shines for interpretable binary classification with linearly separable data, but requires feature independence and linear log-odds relationships to produce reliable results.