Abstract: Transaction records for self-service checkout transactions are provided in real-time to a customizable machine-learning rescan audit service. Any retailer defined settings with respect to random audits are enforced by the service and the real-time transaction records are provided to a machine-learning model that returns rescan scores based on the real-time transaction records. The service compares the rescan scores in view of retailer-defined suspicious scores for both a full rescan of a given checkout transaction and a partial rescan of the given checkout transaction. The service alerts the retailer when a rescan audit is predicted to be warranted for any given self-service checkout transaction based on random selection or based on the computed rescan score. The machine-learning model is continuously and regularly retrained. In an embodiment, the service provides reports, a dashboard, and mining of the transaction records and the audits to the retailer though a service-provided interface.