Analysis and Use of Machine Learning Techniques in Intrusion Detection Systems

In today's technology-oriented world, systems such as public institutions and organizations, healthcare institutions and banks face cyber security risks and threats. The use of Intrusion Detection Systems is very important in ensuring the security of systems and applications on the internet. Intrusion detection systems, which examine the movements occurring on the Internet network, detect abnormal behavior and overcome threats, are used successfully to ensure the security of the system. In addition to Classical Intrusion Detection Systems, it is possible to keep systems up-to-date by detecting and analyzing cyber-attacks with machine learning methods. In this thesis, a comprehensive analysis study was conducted for the performance of a machine learning-based intrusion detection system that can detect malicious activities on the system. CSE-CIC-IDS-2018 data set was used as the data set in this research. While carrying out the study, attributes were selected for each attack by first preprocessing the data for each attack type, and then analyzing the information collected from three different feature selection methods. Then, the performance and performance of STS were evaluated with machine learning techniques and metrics such as accuracy, sensitivity, precision and F-Score and confusion matrix. As a result of the studies, the Random Forest algorithm achieved the best success on average in all attack types, with an accuracy rate of approximately 95%.