Performance Comparison between Principal Component Analysis and Linear Discriminant Analysis in a Gender Recognition System

Abstract

Abstract-In this paper we present a performance comparison of two feature extraction methods applied in a gender recognition system. The gender recognition systems built in this paper consist of pre-processing steps, feature extraction methods, and classifiers. The two feature extraction methods compared in this paper are Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA). The classifier used in the system is nearest neighbor classifier using Euclidean distance measure. The performances of the systems are expressed by their accuracies which are measured using 2-, 5- and 10-fold cross validation. Our experiments show that the best performance is achieved by the system using LDA. The performance of the proposed system is evaluated using two types of data sets, namely real-world data (from which we create the cropped and non cropped data sets) and a controlled environment data from the VISiO-Lab data set. The system achieves an average of accuracies of 81.08% for cropped test data set, 62.66% for non cropped test data set, 85.90% for VISiO-lab data set using 10-fold cross validation