A Performance Analysis of Machine Learning Models for Attack Prediction using Different Feature Selection Techniques

Abstract

The cyber assault has emerged as a serious digital threat. Day after day, hackers inflict severe financial harm on nations and people alike. As a result, threat detection has become increasingly critical for computer system security going forward. This research focused on seeing if it can recognize and anticipate dangers based on complicated instances by using machine learning. This can help identify and prevent attacks before they happen. Many well-known threat classification methods are tested in this work. It is centered around six of the most prominent machine learning (ML) techniques used in intrusion detection research: Random Forest, Naive Bayes, Artificial Neural Networks, K-nearest neighbor algorithm, Support Vector Machines, and bagging. In this study, a machine learning-based strategy for reliably predicting dos, R2L, U2R, probe, and overall attacks was established by assessing the accuracy of several ML tactics. Finding effective machine learning algorithms, as well as determining critical qualities that can provide the best outcomes, is a massive task for academics. This paper also examines the benefits and drawbacks of each method before applying it to the NSL-KDD dataset to evaluate how well they perform. The novelty in this research work lies in the fact that it utilizes various feature selection methods to find the set of features that have the greatest impact on classification performance. Analysis of the results using different performance metrics revealed improved accuracy. For the NSL-KDD dataset's intrusion detection and attack type classification tasks, the highest detection rate was obtained with 99.86%.