Machine learning-based prototype pollution attack detection model

JavaScript, a well-recognized Object-Oriented programming language, utilizes prototypes (a mechanism that allows objects to inherit properties and methods of a class) to handle object attributes. This prototype-based language is prone to a prototype pollution attack. Prototype pollution is a security vulnerability that enables malicious users to modify object prototypes. Existing methods to detect such vulnerabilities carry issues related to accuracy, completeness, and their inability to manage vulnerabilities arising from new third-party libraries. This study seeks to address these shortcomings by proposing a machine learning-based detection approach designed to detect prototype pollution attacks. Unlike existing methods, that are limited to specific frameworks or fixed input patterns, this model employs supervised machine learning algorithms to identify attacks across a variety of JavaScript programming environments. The proposed research evaluates supervised machine learning classifiers, including Logistic Regression, Random Forest, Naive Bayes, K-Neighbors, Support Vector Machine, and Decision Tree, to ascertain the most efficient approach for detecting prototype pollution attacks. The experimental findings reveal that K-Neighbors and Logistic Regression demonstrate impressive performance, achieving an accuracy of 98.9% and an F1-score of 99% on the customized dataset. This study highlights the potential of machine learning for detecting prototype pollution attacks on the tested dataset and identifies the pathways for future advancement. However, its effectiveness needs testing on larger and more diverse datasets to demonstrate its applicability.