A Machine Learning Approach for Intelligent Digital Network Traffic Classification Using XG Boost and Random Forest

Abstract

In the last decade, with the expansion of digital networks, the network traffic control problem has raised the necessity of effective and accurate techniques for its management. This work aims to compare and analyze two ensemble machine learning algorithms, XGBoost and Random Forest, for network traffic congestion classification and management. The proposed approach consisted of: data preprocessing, feature selection applying the minimum redundancy maximum relevance criterion, keeping the top 10 features, and separating the training and testing data with a ratio of 80:20. We trained and tested both models in the network traffic dataset and compared them on the basis of accuracy, precision, recall and confusion matrix. The experimental results confirm that XGBoost had the best classification accuracy 98.75%, precision of 0.987 and recall of 0.986 than Random Forest having classification accuracy 96.42%, precision of 0.965 and recall of 0.961. The main contributions of this work were to offer a benchmark of two tree-based ensemble methods for digital network traffic classification and to show that XGBoost would be a promising candidate for real-time intelligent network traffic control systems, and give some indication for network operators in terms of optimized network resource allocation and congestion reduction by means of automated, data-driven decision making.