The recent frequent occurrence of students’ psychological crisis events has drawn widespread attention to mental health education in colleges and universities. Based on students’ behavioral data, we use big data and data mining technology to model and analyze students’ daily behaviors, complete the construction of students’ social intimacy features based on Dijkstra’s algorithm, use the C4.5 decision tree improvement algorithm based on variable-precision rough set to realize the identification of students’ psychological problems, and analyze the intervention paths of students’ psychological problems and the evaluation of the results of the intervention. The proposed method can recognize students’ psychological problems more accurately, and the recognition accuracy of different levels of psychological problems reaches more than 72%, which is significantly higher than other classification methods. Learning anxiety, loneliness tendency and terror tendency of students in the intervention group were significantly reduced after the psychological intervention (P < 0.05), and the overall factor scores decreased by 9.85%, and the level of mental health was answered to be improved, which reflected the effectiveness of the proposed mental health intervention. The experiment proves that the model in this paper can effectively identify students with psychological abnormalities, and the proposed intervention path for students' psychological problems has a positive impact on the development of students' mental health.