New hybrid attribute selection method and its application in intrusion detection

The performance of a detection method is degraded by the high dimensional data containing irrelevant and redundant attributes. A new hybrid attribute selection method integrating filter and wrapper methods is proposed. Filter method based on mutual information is firstly used to remove irrelevant attributes. Wrapper method based on improved adaptive genetic algorithm and improved evaluation function is used to select optimal attribute subset. Applications in intrusion detection showed that this approach can reduce the time of attribute selection and has better performance in terms of true positive rate and false positive rate than other methods.