Development of an Intrusion Detection System (IDS) Using Machine Learning Algorithms

Abstract

As cyber threats evolve, traditional Intrusion Detection Systems face significant problems such as high false positives, inability to detect new attack signatures, and poor scalability. This paper aims to mitigate these limitations by implementing an IDS based on machine learning techniques to improve the detection capability, flexibility, and effectiveness. Exploiting the realistic and diverse CICIDS 2017 dataset that is quite famous for its realistic and complex attack scenarios, the study does an excellent job of including several enhanced preprocessing techniques to have the best feature selection techniques and, hence, the best data quality. First, an experimental comparison of Support Vector Machines (SVMs), Random Forests (RFs), and neural networks' performance in detecting threatening activities is made to choose the most suitable machine-learning approaches. The quantitative performance of each model is compared and assessed using model accuracy, Precision, recall, F1-score, and the ROC-AUC curve. This research illustrates how machine learning can solve cybersecurity problems by presenting an adaptive algorithm that can disregard emerging threats. The observations benefit the ongoing enhancement of IDS and provide an understanding of deploying efficient and progressive solutions in live networks.