<p>The increasing incidence of psychological health challenges among college students highlights the limitations of traditional education and intervention methods. This study proposes and validates an AI-driven framework integrating predictive modeling and personalized interventions to enhance psychological health education. The framework employs a hybrid model combining LSTM, BiLSTM, and LightGBM to analyze multimodal academic, physiological, and speech-text data for predicting psychological health risk levels. An intelligent intervention system delivers real-time, tailored support through sentiment analysis and recommendation algorithms. The evaluation of 12,543 anonymous student records showed high predictive performance, with a recall rate of 0.93 for high-risk students, an F1-score of 0.91, and a macro average F1-score of 0.85. The results showed that the average academic performance of high-risk and low-risk students was 65.2 and 82.4 points, respectively. AI-driven interventions significantly improved academic results, particularly for moderate-risk students, showing an average gain of 5.1 points. These findings confirm the framework's efficacy in early risk identification and timely intervention, offering a scalable paradigm to improve mental well-being and academic success in higher education.</p>