Deep Learning within the Web Application Security Scope – Literature Review

Abstract

Over the last few years, several breakthroughs in deep learning have contributed to the development of new models. One of many areas they are applied to is the web application security scope. Web applications are still one of the biggest information and business security threats. Requests sent to the Web application are divided into normal and malicious. Malicious requests contain a payload that exploits a discovered vulnerability. Detection of Web attacks can be reduced to natural language processing classification problem. Lately, pre-trained models on Transformer neural networks showed promising results in the detection of Web attacks. In development of models, the preprocessing step of data preparation is crucial. After preparation of good datasets and application of powerful models it is very important to evaluate and compare performance of algorithms. The goal of this paper is to conduct an overview of the deep learning methods used for Web attack detection. The research is conducted by querying scientific databases, analyzing relevant articles within the security scope, and summarizing the proposed state-of-the-art approaches. Findings of reviewed papers were summarized based on implementation details and used performance metrics. Also, open problems will be emphasized, as well as challenges and possibly new opportunities for the future research.