Adaptive feedback in puzzle game boosts ML problem-solving in 205 students

A new study from Clemens Witt and colleagues at WiPSCE 2026 tackles a core challenge in machine learning education: helping students move beyond trial-and-error. The researchers augmented a digital puzzle game for decision tree construction with an adaptive feedback module that generates individualized messages based on continuous evaluation of each learner's problem-solving strategy. This approach targets metacognitive regulation, the key difficulty that causes students to abandon structured approaches in favor of exploratory behavior.

Using a robust dataset of 205 students, including approximately 55 hours of screencast video and detailed gameplay logs, the team analyzed strategic behavior, persistence, and transitions between problem-solving modes. Their findings demonstrate that strategy-oriented feedback effectively supports the development of structured problem-solving skills in decision tree construction. The paper, published in the ACM International Conference Proceedings Series, provides concrete design guidelines for ML learning environments that foster transferable competencies in secondary computing education, moving beyond rote task completion.