Based on the concept of OBE, this paper explores the data-driven learning behavior of preschool music students. A multi-dimensional data analysis framework based on learning engagement was constructed, covering four types of indicators: behavioral, cognitive, emotional, and social. Through the whole-process behavior log and multi-modal interaction data of the MOOC platform, combined with the logistic regression model and social network analysis method, the learner’s input state and evolution law are dynamically identified. The empirical results show that the model has a good prediction effect on the shallow, middle and deep inputs, and the average F1 values of the molecular sets of shallow, middle and deep inputs are all above 0.6. The 250 students were divided into three types of prototypes, with Prototype 1 having a medium overall level of engagement (3.20 points), Prototype 2 having an overall level of medium to high (3.98 points), and Prototype 3 having a high overall level of engagement (4.59 points). In terms of behavioral engagement, there was the largest difference in the number of “mutual evaluations” (standard deviation of 32.58), the difference in the quality value of peer comments and the entropy of self-rated information under different learning themes was large in cognitive engagement, the proportion of positive comments in online learning was more in emotional engagement (mean was 0.62), the overall network density difference was small (standard deviation was 0.06), and the individual network difference was larger. The results of this study provide data-driven decision support for the precise teaching design and personalized intervention of preschool music courses.
