AutoEncoder Based Feature Extraction for Multi-Malicious Traffic Classification

Abstract

In recent years, research is being activated to classify deep learning-based malicious network traffic. Malicious network traffic classification has a problem of wasting time by learning meaningless features due to a large number of traffic and high-dimensional features. In this paper, we propose a technique for feature extraction based on AutoEncoder and classifying malicious network traffic through a random forest classifier. This technique reduces the time and spatial complexity required in the intrusion detection system by extracting features from high-dimensional data. To evaluate this technique, the performance of AE-RF and Single-RF classifiers is measured for Accuracy, Precision, Recall and F-Score using the CICIDS 2017 data set. The evaluation showed that AE-RF has an accuracy of 98% or more, which shows excellent performance and detection speed.