An approach of Bagging ensemble based on feature set and application for traffic classification

Bagging is a classic ensemble approach,whose effectiveness depends on the diversity of component base classifiers.In order to gain the largest diversity,employing genetic algorithms to get independent feature subset for each base classifier was proposed.Meanwhile,for better generalization,the optimal weights for the base classifiers according to their predictive performance were selected.Finally,refined Bagging ensemble based on simple Softmax regression was applied successfully in traffic classification.The experiment result shows that the proposed approach can get more improvement than the original Bagging ensemble in classification performance,and is better than the random-forests to a certain extent.