AI-Driven Post-Game Scrabble Analysis for Rule-Violation Detection and Classification.

Abstract

This study presents an Artificial Intelligence (AI)-based computer vision approach for automated post-game detection of rule violations in physical Scrabble gameplay. Manual identification of incorrect tile placements can be time-consuming and dependent on player or expert judgment, creating a need for an automated method that can analyze board images consistently. The proposed approach uses a YOLO-based object detection model to identify and localize rule-violating tile placements on a standard 15 × 15 Scrabble board. A custom dataset of 1,500 annotated images was developed, containing five categories of rule violations: incorrect initialization, insufficient tile count, non-linear placement, lack of connectivity, and invalid adjacency. Each violation category was treated as a separate detection class, with bounding-box annotations used to identify the locations of incorrect placements. YOLOv8, YOLOv9, and YOLOv10 models were trained and experimentally compared to determine the most suitable model for rule-violation detection. Among the evaluated models, YOLOv9 achieved the strongest performance, obtaining a precision of 97.99%, recall of 98.51%, mAP@50 of 99.00%, and mAP@50–95 of 91.21%. These results demonstrate the effectiveness of the YOLO-based approach in detecting and localizing predefined Scrabble rule violations from board images. The proposed system provides visually interpretable post-game feedback through bounding boxes, supporting automated and consistent review of rule adherence in physical Scrabble gameplay.