Network Traffic Classification using Long-Short Term Memory Algorithm on UNSWNB15 and KDDCUP99 Data Set

Abstract

Many systems rely on the ability to categorize network traffic for purposes including intrusion detection, policy enforcement, and traffic management. Machine Learning (ML) and specifically Deep Learning (DL) based classifiers have shown excellent performance in network traffic categorization, even though most apps encrypt their network data and some dynamically alter their port numbers. Since network traffic flows may be correlated, this study provides a classification strategy based on graph convolution and Long-Short Term Memory (LSTM). To extract the spatial characteristics of the spatial topology and the temporal aspects of the LSTM, the traffic flow data must first undergo data preprocessing. Finally, the method is tested on a subset of the UNSWNB15 and KDDCUP99 datasets to measure its efficiency. The suggested technique has been shown to extract possible characteristics from network traffic data in experiments successfully. It demonstrates the efficacy of the suggested approach and outperforms alternatives such as feature selection, bidirectional LSTM (BiDLSTM), and CNN-LSTM in classification performance.