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

What this post added

This post details the application of machine learning to Cloudflare's WAF, focusing on overcoming data challenges. It introduces data augmentation and generation techniques to improve model accuracy and reduce false positives. Specific methods discussed include generating heterogeneous negative samples, creating pseudo-random noise samples with increasing complexity, and mutating benign content. The post explains how these techniques help the ML model learn structural, semantic, and statistical properties of content, making it less sensitive to individual malicious tokens and more robust against obfuscation.