Improving the WAF with Machine Learning

Introduces a new machine learning model for WAF detection that scores every request for SQLi and XSS attacks. The model is trained on existing WAF-classified traffic and augmented data. It provides three scores: a total score (`cf.waf.ml.score`), and specific scores for SQLi (`cf.waf.ml.score.sqli`) and XSS (`cf.waf.ml.score.xss`), ranging from 1 (malicious) to 99 (valid). These scores can be used in custom WAF/Firewall rules for mitigation. The system aims to identify bypasses and anomalies proactively and runs on all traffic with negligible latency. The post details the challenges in building a reliable dataset, scalable data labeling, model selection, and high-throughput execution.