Feature Reduction to Classifiers

This case study revealed the influence of 4 common feature reduction methods (PCA, LDA, and their kernel versions) on the 4 different types of classifiers (SVM, ML, KNN, GMM). Our experiments showed that SVM was the most robust method to increase dimensional space, and that SVM and LDA were the most sensitive to noise.

Keywords: Feature reduction, Classification, PCA, LDA, Kernels, SVM, ML, KNN, GMM, High Dimension

• Course Project of “Digital Pattern Recognition”
• Instructor: Dr. Saurabh Prasad
• Report: [Pre-print]
• Code: [Matlab]