Application of Machine Learning Methods for Classification of Telemetric Frames by Compression Algorithms

Abstract

Problem statement: adaptive data compression systems include, as an integral part, a certain classifier that allows you to select the most efficient compression algorithm. At the same time, for such systems, parallel training of the classifier and data compression should be ensured, which increases the computational cost and complicates the architecture of the transceiver devices. Thus, the question of developing an effective classifier for compression systems is quite acute. Objective: to assess the possibility of using the simplest neural network architecture as a classifier for telemetric data frames. Results: the behavior of the averaged errors of training, generalization and confirmation depending on the size of the training sample obtained for a number of telemetry data sets is investigated. Based on the obtained data, neural network training parameters are proposed that allow achieving high efficiency of its work. A comparative analysis of the effectiveness of the proposed approach and a number of specialized numerical methods has been carried out, which has shown that even the most versatile artificial neural network architecture is significantly superior to systems using numerical classification methods.