Currently, there is a shift from one-time academic education to lifelong learning, highlighting the necessity of self-directed learning ability. To cultivate learners with a high level of autonomy, it is essential to predict and diagnose their self-directed learning abilities. Given the limitations of existing diagnostic methods for self-directed learning ability, this paper focuses on students in the software engineering course at Peking University. It constructs a hybrid learning behavior feature on the basis of Zimmerman's self-regulated learning model and characteristic instructional design. A predictive model for self-directed learning ability based on learning behaviors is designed to enhance teaching quality and improve self-directed learning ability.