<p>The early identification and the classification of the psychological issues of students is essential in ensuring academic success, emotional well-being and timely interventions. The approaches based on deep learning provide a viable opportunity as they use behavioral and physiological signals to reveal concealed psychological states. Current techniques are usually subjective and based on questionnaires or fixed characteristics, are subject to subjectivity, delays in responding, and inaccuracy in real-time tracking. Such weaknesses decrease their efficiency in the process of recording delicate, dynamic psychological differences. This research presents a hybrid deep learning system that is suggested to mitigate these issues, namely the Psycho-Behavioral Temporal Deep Network (PBTDN). PBTDN combines multimodal information sources facial micro-expression recognition, voice stress patterns and activity-related logs into a time-based CNN-BiLSTM pipeline. The convolutional layers are used to learn local discriminative features and Bidirectional Long Short-Term Memory (BiLSTM) layers to learn temporal dynamics to ensure strong recognition of changing psychological conditions. The given approach can be utilized in the learning environment to track the students in a non-invasive way so that counselors and teachers could detect the initial symptoms of stress, anxiety, or depression. These real-time, data-driven insights promote proactive interventions and personalized support. The experimental results show that PBTDN has better performance compared to traditional methods as it is more accurate, sensitive and able to provide time stability in the detection and classification of psychological issues among students.</p>