Linear Discriminant Analysis - niranjv/ml-notes GitHub Wiki

Overview

  • Logistic regression models the conditional distribution P(Y = k|X = x)
  • Alternatively, model distribution of X separately in each class and use Bayes theorem to calculate P(Y = k|X = x). This is similar to logistic regression if the distributions are Normal
  • This alternative method when:
    • Classes are well separated => parameter estimates for logistic regression model are unstable (does not happen in LDA)
    • For small n and Normal distribution of X, LDA is more stable than logistic regression
    • LDA works well when |classes| > 2

Ref

  • ISLR, Section 4.4
⚠️ **GitHub.com Fallback** ⚠️