Özet:

The increasing use of Internet-based technologies has brought significant advantages to people and organizations as well as a number of disadvantages. The most important of these are cyber attacks. With the diversification and increase of cyber attacks, the protection of critical data that reach large amounts against actions such as deletion, modification, disclosure becomes more difficult every day. Therefore, the attack detection systems, which are one of the tools developed for the purpose of ensuring the security of information systems, have become a very important work area. In this study, the CSE-CIC-IDS2018 data set, using the various preference selection methods and machine learning techniques suggested in literature, examined the impact of preference selection on the success and performance of the Attack Detection System. To determine the best subgroup that can represent the original data set, the Ki-Care Test, Spearman's Ranking Correlation Scoring and Self-Animated Properties Elimination methods have been used. A comparative analysis of performance results has been made by classifying the new data sets by the methods of Adaptive Enhancement, Decision Tree, Logistic Regression, Multi-layer Sensor, Extra Tree, Passive-Agrictive and Gradyan Enhancement Machine Learning. K-Fold is used to evaluate performance objectively. The K-Fold process is expensive in terms of calculation and time, and the processing time is reduced by the application of paralysis. According to the experimental results obtained by Ki-Kare Test and Spearman’s Ranking Correlation Score, the preferential selection methods reduced the processing load by 45% due to the reduction of data size, but increased the error rate by 14.46% and 10.52%, respectively. Furthermore, when the self-release proprietary elimination method is used with appropriate settings parameters, it has been shown that the system has reduced the error rate to 2.95% with a reduction of the processing time by 38%.

Anahtar Kelimeler: