Machine learning model and mood fluctuation prediction for college students' mental health intervention

Aiming at the increasingly prominent situation of college students' mental health problems, this study puts forward a prediction and intervention model of college students' emotional fluctuation based on machine learning. By integrating multi-source heterogeneous data such as text, physiological signals and time characteristics, a hybrid model integrating BiLSTM-Attention-CNN is constructed to improve the accuracy of short-term mood fluctuation prediction. The experimental results show that the model achieves an accuracy of 82.3%, a F1 score of 0.74 (for high-risk categories) and an AUC-ROC value of 0.85 in the task of mood fluctuation prediction, which is superior to the traditional machine learning model and the monomodal model. Based on the predicted results, a dynamic intervention framework combined with the principle of cognitive behavioral therapy (CBT) is designed to provide personalized intervention programs for students with different risk levels. In the 4-week dynamic intervention test on 60 high-risk students, the PHQ-9 score decreased significantly. The score of the high-risk group decreased from 23.4 ± 2.1 to 16.8 ± 3.2, a decrease of 28.2%, and the score of the medium risk group decreased from 15.1 ± 1.8 to 11.3 ± 2.4, a decrease of 25.2%. The results of the two groups were statistically significant (p<0.05). This study provides a theoretical basis and technical reference for the intelligent upgrading of mental health services in Colleges and universities, helps to realize the transformation from "passive response" to "active prevention", and provides a new and effective way for college students' mental health intervention.