THE EFFECT OF MACHINE LEARNING ON INTRUSION DETECTION SYSTEMS

As technology advances and the link between people and machines grows, system and data security become more important. Attackers try to find gaps by examining systems and sometimes succeed. Successful attacks lead to material and moral damages. Anti-virus or firewalls are used to prevent them. Anti-virus and firewalls may not always provide an effective defense against expert attackers. Based on these and similar problems, intrusion detection systems have been developed. They do this by collecting information from various systems and network resources and then analyzing the data for possible security issues. This study focuses on these problems and aims to train an intrusion detection system using machine learning techniques, known attack types, and data from server-based attack methods. In this direction, the data set was created by combining CesarFTP, WebDAV, Icecast, Tomcat, OS SMB, OS Print Spool, PMWiki, Wireless Karma, PDF N, Backdoored Executable, Browser Attack, Infectious Media attack data. The resulting data set was classified and trained using the Support Vector Machine (DVM) and Naive Bayes (NB), and the results were shared. Following the training and testing of the system with DVM, the success rate of 0.7129 was achieved, followed by the re-applied size reduction and Principal Component Analysis with Naive Bayes and the success level of 0.7914. This showed that the intrusion detection system, which was trained using the aforementioned intrusion data, was able to detect 79 percent of incoming attacks accurately while it was active and operational