Framework-based Roguelike Game for AI/ML Education

The paper introduces RGAME, a roguelike serious game designed to teach the machine learning workflow (data collection, preprocessing, training, and evaluation) to Japanese K-12 students, specifically middle and high school learners aged 13 to 18. Its central premise is that most existing AI/ML education games emphasize coding rather than conceptual understanding and have seen limited adoption or evaluation. RGAME addresses this by pairing a roguelike game, which uses randomized environments and permanent death to support iterative play, with Google Colaboratory, where players run ML code using scikit-learn models including linear SVC, KNN, SVC, and XGBoost.

The game's design is grounded in self-determination theory and in established frameworks for educational games and AI literacy. Game elements such as points, badges, performance graphs, and meaningful stories are mapped to motivators and psychological needs for competence, autonomy, and relatedness. Model accuracy on test data determines the player's attack power, providing regular feedback. The authors also draw on 15 AI education design considerations to justify the tool choices, though the paper presents these mappings conceptually rather than through empirical results.

The study is at a proposal and planned-evaluation stage. No experimental data are reported. The authors outline a comparison between a full game version (FUL) and an e-learning version without game elements (ELE), using pre-tests, post-tests, and post-engagement surveys administered online via Google Forms. Assessments are to be built on the Five Big Ideas of AI, while the survey adapts the Serious Games Evaluation Scale and the Ubisoft Perceived Experience Questionnaire. The paper therefore establishes a design rationale and evaluation plan, and it anticipates positive outcomes without yet demonstrating them.