Overview of public datasets for web application attack detecting

Abstract

Currently, web applications are one of the most popular means of providing services to the end user. Protection against attacks on web applications only increases its relevance every year. Modern Web Application Firewalls (WAF) often use machine learning in one form or another. Recent research shows that systems for detecting malicious HTTP requests using classical machine learning and deep learning in most cases outperform systems based on explicitly prescribed rules in terms of efficiency. Nevertheless, there are problems with reproducibility of experiments described in publications on this topic. In most works, there are results of evaluating proposed models based on publicly unavailable datasets; the result of the model's work on several publicly available datasets is not considered. This paper provides an overview of open (publicly available) datasets on the basis of which models for detecting malicious web requests can be trained and evaluated by metrics. This overview includes both widely used datasets for benchmarks and less well-known ones. In addition, the overview can be useful when forming combined datasets.