As one of the most desired skills for contemporary education and career, problem-solving is fundamental and critical in game-based learning research. However, students' implicit and self-controlled learning processes in games make it difficult to understand their problem-solving behaviors. Observational and qualitative methods, such as interviews and exams, fail to capture students' in-process difficulties. By integrating data mining techniques, this study explored students' problem-solving processes in a puzzle-based game. First, we applied the Continuous Hidden Markov Model to identify students' problem-solving phases and the transition probabilities between these phases. Second, we employed sequence mining techniques to investigate problem-solving patterns and strategies facilitating students' problem-solving processes. The results suggested that most students were stuck in certain phases, with only a few able to transfer to systematic phases by applying efficient strategies. At the beginning of the puzzle, the most popular strategy was testing one dimension of the solution at each attempt. In contrast, the other two strategies (remove or add untested dimensions one by one) played pivotal roles in promoting transitions to higher problem-solving phases. The findings of this study shed light on when, how, and why students advanced their effective problem-solving processes. Using the Continuous Hidden Markov Model and sequence mining techniques, we provide considerable promise for uncovering students' problem-solving processes, which helps trigger future scaffolds and interventions to support students’ personalized learning in game-based learning environments.
Scopus Subject Areas
- Computer Science(all)
- Human-computer interface
- Data science applications in education