XSS Attack Detection with N-Gram Based Prediction Model

The increment developments in technology has empowered the web applications. Meanwhile, the existence of Cross-Site Scripting (XSS) vulnerabilities in web applications has become a concern for users. In spite of the numerous current detection approaches, attackers have been exploiting XSS vulnerabilities for years, causing harm to the internet users. In this paper, a text-mining based approach to detect XSS attacks in web applications is introduced. This approach is built to extract a set of features from a publicly available source code files, which are then used to build a prediction model. The findings include few comparisons between Word Tokenization and N-Gram in accuracy, time spend to build the model and AUC-ROC curve. The results show that N-Gram tokenization outperforms the Word Tokenization.