Refer to the exhibit.


A FortiWeb administrator tests a new form input value after training the machine learning (ML) anomaly detection system.
The hidden Markov model (HMM) flags the input as abnormal, while the support vector machine (SVM) model classifies it as normal. FortiWeb allows the request.
What does this result indicate about the FortiWeb ML anomaly detection behavior?
FortiWeb machine learning uses layered detection rather than treating every unusual value as malicious. The HMM layer models normal parameter behavior and can flag a value as abnormal when it falls outside the learned distribution. However, abnormal does not automatically mean hostile. FortiWeb then uses additional ML classification logic, including SVM-based evaluation, to determine whether the anomaly resembles an actual attack or simply a legitimate unusual input. In this case, HMM noticed that the value was uncommon, but SVM classified it as normal, so FortiWeb allowed the request. That is expected behavior. Raising thresholds, disabling models, or assuming FortiWeb failed would misunderstand the two-stage ML decision process.
