Saldırı Tespit Sistemlerinde Makine Öğrenmesi Modellerinin Karşılaştırılması

Abstract

As a result of developing technologies in recent years, all kinds of computing devices can be connected to the Internet. In this way, many real-world problems are transferred to the new network layout, but this uncontrollable virtual platform contains many vulnerabilities. One task of network administrators is closing these leaks and protecting the network from attacks. Although use of firewalls can prevent serious attacks from outside, there are many attacks from inside or previously unknown. Intrusion Detection Systems (IDSs) are the most preferable applications to eliminate these vulnerabilities. When recently IDSs are examined, it is seen that Machine Learning-based systems are focused on in order to develop a dynamic security mechanism. It is seen that developments in hardware and parallel computing and Big Data processing technologies are used compatible with these systems. In this study, it is aimed to develop STS using seven different algorithms. Results were compared in terms of performance, training and running times, and appropriate algorithm was determined. NSL-KDD dataset was used as generally accepted-dataset. The results showed Adaboost algorithm achieves the highest accuracy. However, when both training-time and runtime performance are considered, Decision Tree algorithm performs better and close to Adaboost in terms of accuracy.