Claim-Level Denial Prediction

Predictive Denial Model

Trains a per-claim Naive Bayes model on the CCD, scores the held-out split, and flags systematic misses — claims the model predicts should be denied but were paid (recoverable revenue opportunity) and systematic false positives — denied claims the model thinks should have paid. Feeds the EBITDA bridge denial-reduction lever with a data-driven target instead of an industry aggregate.