What Is Linear Discriminant Analysis?

Linear Discriminant Analysis Explainer

Linear discriminant analysis, or LDA, is a supervised method for classification and dimensionality reduction. It models each class distribution and finds linear combinations of features that help separate class centers relative to variation within classes.

The standard classifier assumes class-conditional Gaussian distributions with a shared covariance matrix. Class priors and estimated means then determine linear decision boundaries and posterior class probabilities.

For dimensionality reduction, LDA projects data into directions chosen using class labels. That distinguishes it from principal component analysis, which seeks directions of overall variance without using the target classes.

Performance can suffer when covariance estimates are unstable, classes are strongly non-Gaussian, or boundaries are nonlinear. Scaling, regularization, sample size, class imbalance, and honest validation should guide whether LDA is appropriate.

Linear discriminant analysis separates classes with linear combinations of features under assumptions about class distributions and covariance.

Key terms include linear discriminant analysis, class mean, covariance, class prior, decision boundary, supervised projection and posterior probability.

LDA chooses linear directions that make class centers distinct relative to variation inside each class.

The classifier combines estimated distributions and class priors to calculate linear boundaries.

Shared covariance and distribution assumptions make the model efficient and interpretable.

Regularization or a different classifier may be needed when data are limited or the assumptions fit poorly.

Explore more "Explainers"

Discover additional explainers across politics, science, business, technology, and other fields. Each explainer breaks down a complex idea into clear, everyday language—helping you better understand how major concepts, systems, and debates shape the world around us.