Improving the accuracy of our machine learning WAF using data augmentation and sampling

This post details the challenges in acquiring high-quality labeled data for machine learning-based WAFs, including privacy constraints, heterogeneity of samples, sample difficulty, and noisy labels. It introduces data augmentation as a strategy to overcome these challenges, focusing on generating heterogeneous negative samples by mutating benign content and creating pseudo-random noise samples with increasing complexity. This approach aims to reduce false positives and improve the robustness and overall properties of the ML WAF.