Abstract: Under the memetic framework， a new feature selection method combining filter and wrapper models was proposed. In the hybrid algorithm， classifier accruracy was used as fitness function to ensure global optimization， while joint mutual information was used as evaluation indicator to accelerate the process. The experimental results indicated that the proposed method outperformed the existing methods in computational efficeicecy and number of selected features. Applying this algorithm to traffic classification resulted in the improved accuracy with fewer features.

Key words: mutual information, memetic framework algorithm, feature selection, traffic classification