Ridge-forward quadratic discriminant analysis in high-dimensional situations

Quadratic discriminant analysis is a classical and popular classification tool, but it fails to work in high-dimensional situations where the dimension p is larger than the sample size n. To address this issue, the authors propose a ridge-forward quadratic discriminant (RFQD) analysis method via screening relevant predictors in a successive manner to reduce misclassification rate. The authors use extended Bayesian information criterion to determine the final model and prove that RFQD is selection consistent. Monte Carlo simulations are conducted to examine its performance.