<p>Depression among students has become increasing concern as social media gradually shapes their everyday involvements, emotional well-being, and personal communications. While many studies attempt to examine online behavior for early mental-health valuation, current methods frequently struggle to detect complex patterns in user activity and offer limited predictive accuracy. This study proposes Hy-OCNN-BiLSTM, an enhanced deep-learning–based framework that uses students’ social-media usage patterns to estimate their risk of depression. The proposed method combines a hybrid neural-network architecture with an optimization-driven feature-selection strategy to better represent behavioral signals associated with emotional distress. The proposed framework integrates convolutional learning for pattern extraction with bidirectional sequence modeling for behavioral interpretation. Model parameters and feature relevance are further refined using modern optimization and feature-selection strategies to improve generalization and interpretability. Experimental results on publicly available dataset demonstrate that the proposed approach achieves substantially higher accuracy score of 0.9968 than baseline models, demonstrating its effectiveness in detecting early signs of depression from social-media behavior. These findings highlight the potential of data-driven systems to support early mental-health screening and to inform preventive interventions for student populations.</p>