The core problem tackled by this research is the opaque nature of many high-performing AI models used in fraud detection.