Examining self-regulation models of programming students in visual environments: A bottom-up analysis of learning behaviour

Self-regulated learning (SRL) significantly impacts the process and outcome of programming problem-solving. Studies on SRL behavioural patterns of programming students based on trace data are limited in number and lack of coverage. In this study, hence, the Hidden Markov Model (HMM) was employed to probabilistically mine trace data from a visual programming learning platform, intending to unveil students’ SRL states and patterns during programming problem-solving in a bottom-up manner. Furthermore, the K-means clustering technique was utilized to cluster the Online Self-regulated Learning Questionnaire (OSLQ) survey data, enabling the investigation of prominent behavioural characteristics and patterns among students with differing levels of SRL. The results show that programming problem-solving involves five SRL states: problem information processing, task decomposition and planning, goal-oriented knowledge reconstruction, data modelling and solution formulating. Students with a high level of SRL are more engaged in the problem information processing stage, where they plan task objectives and develop problem-solving strategies by profoundly analyzing the structural relationships of the problem. In contrast, students with low levels of SRL decompose the problem and develop a strategic approach through interacting with the knowledge content, which results in a certain blindness in the problem-solving process.