Measuring the Security Effectiveness of Machine Learning Methods Used Against Cyber Attacks in Web Applications

Abstract

The rapid progress of technological developments in the global world, the people to closely follow these developments and share them have become the focus of cybercriminals. People realize their basic needs, requests, shares or works via smart devices using the internet infrastructure. While performing these actions, users can inevitably leave an open door through web applications. As a result, user-defined information can easily be passed on to others. Recently, there has been a serious increase in activities carried out on websites. One of the reasons for this increase, and the most important one, is the pandemic that has had an impact worldwide. Cybercriminals want to turn such situations into opportunities and gain financial gain. They look for vulnerabilities in the websites that people demand heavily and they want to access their user information and card information. This study proposes an approach that measures the performance of machine learning methods against the vulnerabilities of various websites. The data set used in the study consists of parameter properties of 1000 websites. In the experimental analysis of the study; Multilayer Perceptron, Support Vector Machines, Decision Trees, Naive Bayesian, Random Forest methods were used. The general accuracy achievements obtained from machine learning methods are; it was 74%, 73.7%, 100%, 69.5% and 100%, respectively. Experimental analysis has shown that machine learning methods are effective in detecting cyber attacks.