Efficient Detection of XSS and DDoS Attacks with Bent Functions

In this paper, we investigate the use of Bent functions, particularly the Maiorana–McFarland (M–M) construction, as a nonlinear preprocessing method to enhance machine learning-based detection systems for Distributed Denial of Service (DDoS) and Cross-Site Scripting (XSS) attacks. Experimental results demonstrated consistent improvements in classification performance following the M–M Bent transformation. In labeled DDoS data, classification performance was maintained at 100% accuracy, with improved Kappa statistics and lower misclassification rates. In labeled XSS data, classification accuracy was reduced from 100% to 87.19% to reduce overfitting. The transformed classifier also mitigated overfitting by increasing feature diversity. In DDoS and XSS unlabeled data, accuracy improved from 99.85% to 99.92% in unsupervised learning cases for DDoS, and accuracy improved from 98.94% to 100% in unsupervised learning cases for XSS, with improved cluster separation also being noted. In summary, the results suggest that Bent functions significantly improve DDoS and XSS detection by enhancing the separation of benign and malicious traffic. All of these aspects, along with increased dataset quality, increase our confidence in resilience detection in a cyber detection pipeline.