Evaluation of Supervised Machine Learning Models with Recursive Feature Elimination for Network Traffic Attack Classification

Abstract

The Internet is facing numerous attacks of different kinds that put its data at risk. The safety of information within the network is, therefore, a significant concern. One of the key challenges of machine learning Approaches for Network Traffic Attack Classification is the expensive computational complexity, which is largely due to redundant, incomplete, and irrelevant features contained in datasets for Network Traffic Attack Classification. In this work, we propose an approach for Network Traffic Attack Classification modeling approach with a Recursive Feature Elimination (RFE) algorithm for Feature Selection (FS) in Cognitive Radio Networks. The FS algorithm is a wrapper-based algorithm with a decision tree as the feature evaluator. The proposed FS method is used in combination with some selected supervised Machine Learning algorithms to build Network Traffic Attack Classification models using the CIC-IDS 2017 dataset. We evaluate the effectiveness of our proposed method by comparing the classification performance of different supervised learning algorithms using standard performance metrics. The implemented experiments compare the results of each algorithm and demonstrate that the Random Forest is the best algorithm used for the network traffic classification with accuracy, precision, recall and F1- score parameter of 0.9998, 0.9894, 0.9975 and 0.9934 respectively while Naïve Bayes achieves the lowest accuracy precision, recall and F1- score parameter of 0.9547, 0.7368, 0.9820 and 0.7844 respectively.