A stable feature selection approach for optimizing traffic classification based on adaptive threshold

In recent years, machine learning algorithm has been widely studied in the field of traffic classification. However, most studies focus on performance improvement of classifier, pro-phase work of traffic classification - feature selection is ignored. Therefore, WSU is regarded as metric, an ATFS algorithm - (Adaptive threshold feature select) is designed on the basis. Namely, algorithm is based on precision autonomous selection threshold of classifier aiming at different datasets. Each dataset will generate a set of attribute subset eventually. Stable features are selected in different screened attribute subsets through TRF algorithm, thereby reaching the purpose of high precision. The experiment shows that the features finally selected in the algorithm can reach the precision of > 96% on C4.5 classifier.