AGMV Approach for Reduce Complexity of Classification Tasks

We present a research of applying a statistical method of data processing and generation of a new features for Accumulated Generalized Mean Values (AGMV) for internet traffic analysis. The approach is based on the accumulation of averaged values and application of fractional powers at various scales. We tested “classical” Machine Learning (ML) algorithms for solving classification tasks and show quality improvement. The method was applied on several datasets for investigation the possibilities and to prove the effectiveness of the AGMV: encrypted and unencrypted traffic ISCXVPN (2016) dataset and Android Malware dataset CCCS-CICAndMal (2020). We demonstrate that AGMV preprocessing method significantly improves the performance of ML algorithms. Source code is stored at github. com/AleksandrIvchenko/agmv.