HMM-Based Framework for Proactive Web Application Vulnerability Detection
Vulnerability detection in web applications is essential for protecting data and to ensure safe and secure user interactions. Traditional machine learning (ML) algorithms have achieved significant effectiveness in identifying known vulnerabilities, such as SQL injection (SQLi), cross-site scripting (XSS), and cross-site request forgery (CSRF). However, since their effectiveness often depends on large labelled datasets and static input parameters, they are less efficient in dynamic and changing threat environments. This study addresses the use of Hidden Markov Models (HMMs) as a probabilistic method to forecast sequential user behavior and spot anomalies that could potentially identify web-based vulnerabilities. By capturing the temporal structure of online requests, the proposed HMM architecture identifies deviation patterns that signal malicious behavior. We compare the HMM-based model against prominent machine learning tools like decision trees, SVMs, and ensemble-based scanners like Burp Suite and OWASP ZAP. Experimental results demonstrate the fact that HMMs can achieve competitive True Positive Rates (TPR) and Precision, and that their generative nature makes them more robust to zero-day attacks. We also evaluate key performance metrics which comprises Accuracy, F1 Score, and False Positive Rate (FPR) with the objective to provide a comprehensive performance comparison. The findings show that while ML models excel in classification tasks, HMMs offer an intriguing advantage in detecting sequential and context-sensitive vulnerabilities. Potential improvements to hybridize HMM with ML and increase detection capabilities across heterogeneous web frameworks have been addressed in the paper's conclusion.