We ensured the target models achieved an average performance exceeding 85%, as shown in the following Table.
Table: Detection performance of the ML-based AMD methods.