Two Feature Selection Methods Comparison Chi-square and Relief-F for Facial Expression Recognition
Mayyadah Ramiz Mahmood
Abstract
Abstract Feature selection metho represents one of the main keys that has direct influence on classification accuracy. During the last two decades, researchers have given a lot of attention in feature selection approaches due to their importance. This paper provides a comparative approach between the two feature selection methods: Chi-Square and Relief-F. The two methods rank the features according to their score. The first highest six emotion features from the both methods are selected. The six features are used to compare the accuracy ratio among the four classifiers: Support Vector Machine, K-Nearest, Decision Tree, and Radial Base Function. These classifiers are used for the mission of expression recognition and to compare their proportional performance. The ultimate aim of the provided approach is to use minimum number of features from the both methods in order to distinguish the performance accuracy of the four classifiers. The provided approach has been applied on CK+ facial expression recognition dataset. The result of the experiment illustrates that K-Nearest Neighbor is the most accurate classifier on the both feature selection methods according to the employed dataset. The K-Nearest Neighbor accuracy average rate for Chi-square is 94.18% and for Relief-F is 94.93%.