Cyber Intrusion Detection System based on Machine Learning Classification Approaches

Abstract

As the internet has advanced over the years, so has the number of cyber-attacks. A sophisticated Intrusion Detection System (IDS) is essential to protect the cyberspace. The goal of IDS is to monitor and evaluate the operations that occur in a network for any signals of probable abnormalities. Although little research has been done in this area, more comprehensive research has yet to be completed. By examining the combinations of most prominent feature extraction (FE) techniques and classifiers, this research offers an IDS for networks based on machine learning (ML) that has a good union of FE techniques and classifiers. This paper introduced a feature extraction (FE) approach for classification issues, using independent component analysis (ICA). We can generate new features independent of each other by utilizing ICA to solve supervised classification issues, and we can also accurately express the output information. A set of significant features is selected from the original collection of features using FE algorithms. The set of significant features is then used to train various types of classifiers to produce the IDS. The proposed methods were evaluated in terms of five different performance measures using the DARPA KDD 99. Finally, it is discovered that the proposed ICA+RF classifier outperforms the others with an accuracy of 99.6%, f-score of 92.6%, and false alarm rate (FAR) value of 0.0029. The result was further compared with state-of-the-art, and it was deduced that our system performed better with higher accuracy and lower FAR.