The psychological health of the students is a very high influencer of emotional stability, behavior and academic performance. The crises should be avoided by detecting the anomalies early. The existing methods face the problem of noisy high-dimensional campus data and fail to pick subtle signs of distress in normal variability. Another model that can help solve these issues is the Temporal Sensitive Network (TSN), a new behavioral time series analysis model that will be used to identify psychological distress. TSN uses a two‑phase pipeline. The initial step involves Jenks natural breaks which are applied to the features to give the discretization of the features and then Apriori is used to mine rules that correlate with the health indicators. This step derives discriminative signals on consumption, internet and activity logs. The second phase uses an attention enhanced gated module that combines long-term habits with short-term variations with more preference to anomalies through soft-max-weighted representations. The outcome is a contextual anomaly detector that is dynamic. Experiments on the Student Life data demonstrate that TSN performs better than the usual baseline (RF, SVM, LSTM, ST -GCN) with the accuracy, precision, recall, and F1 score of 78.4, 77.6, 78.0 and 77.8 respectively. The research of Ablation validates the contributions of every constituent, and interpretability visualizations clarify the decision-making of the model. As shown in this work, a privacy preserving, scalable methodology can be used to facilitate proactive campus support systems.