Suggesting the likely category of an exception — timing, fee, duplicate, missing entry, shortage, chargeback — so the investigator starts from a hypothesis rather than a blank page.
Supervised multi-class classification over structured features, with feature attribution for explainability
scikit-learn RandomForestClassifier or GradientBoosting; XGBoost/LightGBM at scale; SHAP for per-prediction explanation