Saldırı Tespit Sistemlerinde Topluluk Öğrenme Yöntemlerinin Kıyaslanması

Abstract

The development of technology in today's information age has led to the formation of various security vulnerabilities. This situation has led to an increase in cyber attacks against individuals, companies and states. Various methods, techniques and commands have been developed to prevent attacks and security vulnerabilities. These structures developed to ensure security are obliged to protect the personal data of users. However, as soon as the attackers detect the security vulnerability with the attack methods what they use, they attack the relevant network and affect the functionality of the network, reducing its performance. Therefore, Intrusion Detection Systems have been developed to secure systems and detect attacks. The use of machine learning algorithms in Intrusion Detection Systems is increasing. In this study, ensemble learning algorithms, Random Forest, CatBoost, XGBoost and LightGBM Intrusion Detection Systems are introduced and compared on NSL-KDD and UNSW-NB15 datasets, which are widely used in anomaly detection. The performances of the algorithms were calculated using the accuracy, precision, recall, f-measure and under-curve performance metrics. In the experiments carried out, the best performance values in both datasets were obtained with the Random Forest algorithm.