A Review on Features’ Robustness in High Diversity Mobile Traffic Classifications

Abstract

Mobile traffics are becoming more dominant due to growing usage of mobile devices and proliferation of IoT. The influx of mobile traffics introduce some new challenges in traffic classifications; namely the diversity complexity and behavioral dynamism complexity. Existing traffic classifications methods are designed for classifying standard protocols and user applications with more deterministic behaviors in small diversity. Currently, flow statistics, payload signature and heuristic traffic attributes are some of the most effective features used to discriminate traffic classes. In this paper, we investigate the correlations of these features to the less-deterministic user application traffic classes based on corresponding classification accuracy. Then, we evaluate the impact of large-scale classification on feature's robustness based on sign of diminishing accuracy. Our experimental results consolidate the needs for unsupervised feature learning to address the dynamism of mobile application behavioral traits for accurate classification on rapidly growing mobile traffics.