A mobile app traffic statistics feature selection method

Abstract

The invention discloses a method for selecting mobile App traffic statistics features, the method comprising: S1, acquiring the original mobile App traffic data set, extracting the flow statistical features of Mobile App traffic, acquiring a marked data set LD for training, and The unlabeled data set UD to be classified; S2. On the LD data set, use the information gain rate to evaluate the ability of each flow statistical feature to distinguish between classes; S3. On the LD and UD data sets, calculate each flow statistics The value distribution of the feature, use the Hellinger distance to evaluate the difference of the feature value distribution, and evaluate the drift degree of the flow statistical feature; S4, use the drift degree as the penalty factor of the distinguishing ability, and calculate the comprehensive evaluation value of the flow statistical feature; S5, based on the comprehensive Evaluation value, searching for a subset of flow statistical features with strong discriminative power and low degree of drift. The method of the invention relates to mobile App traffic classification technology in the field of mobile Internet traffic measurement, reduces data dimension, and improves classification robustness.

Description

technical field [0001] The invention relates to the technical field of traffic classification in the field of traffic measurement, in particular to a method for selecting mobile App traffic statistics features. Background technique [0002] The mobile app traffic classification technology based on machine learning is described as follows: the original traffic is grouped based on the five-tuple {source IP, destination IP, source port, destination port, transport layer protocol} group flow, and the traffic is extracted from the statistical characteristics of the flow to establish a flow sample set. for training classification models. A variety of flow statistical features have been proposed. For example, Moore et al. proposed 248 flow statistical features in 2005 (A.Moore, D.Zuev, M.Crogan. Discriminators for use in flow-based classification. Queen Mary and Westfield College, Department of ComputerScience, 2005.), including: statistical characteristics of packet size (minimum...

AI Technical Summary

Problems solved by technology

Although this invention eliminates redundant features for network traffic data, it does not have the stability of analyzing features and cannot be used to select a more stable feature subset

Images