A Hybrid Gini PSO-SVM Feature Selection: An Empirical Study of Population Sizes on Different Classifier

Abstract

A performance of anti-spam filter not only depends on the number of features and types of classifier that are used, but it also depends on the other parameter settings. Deriving from previous experiments, we extended our work by investigating the effect of population sizes from our proposed method of feature selection on different learning classifier algorithms using Random Forest, Voting, Decision Tree, Support Vector Machine and Stacking. The experiment was conducted on Ling-Spam email dataset. The results showed that the Decision Tree with the smallest size of population is able to give the best result compared to NB, SVM, RF, stacking and voting.A performance of anti-spam filter not only depends on the number of features and types of classifier that are used, but it also depends on the other parameter settings. Deriving from previous experiments, we extended our work by investigating the effect of population sizes from our proposed method of feature selection on different learning classifier algorithms using Random Forest, Voting, Decision Tree, Support Vector Machine and Stacking. The experiment was conducted on Ling-Spam email dataset. The results showed that the Decision Tree with the smallest size of population is able to give the best result compared to NB, SVM, RF, stacking and voting.