STRESS RECOGNITION THROUGH DEEP LEARNING APPROACHES: DATA, MODELS, AND ACADEMIC ENVIRONMENT APPLICATIONS IN REVIEW

Stress is a critical psychological and physiological condition that adversely affects health, cognitiveperformance, and quality of life. In educational environments, persistent stress among studentsundermines academic achievement, engagement, and overall well-being, necessitating accurate andtimely detection. Conventional stress assessment techniques, though effective in laboratory settings,are constrained by reliance on handcrafted features and limited real-world applicability. Recentadvances in deep learning have enabled automated representation learning and multimodalintegration, leveraging physiological signals (EEG, ECG, GSR), behavioural modalities (facialexpressions, speech, text), and wearable sensor data. This review synthesizes state-of-the-art deeplearning methods for stress detection, with emphasis on facial expression recognition as anonintrusive modality in classroom settings. Architectures including convolutional neural networks(CNNs), recurrent neural networks (RNNs), long short-term memory (LSTM) models, attentionmechanisms, and hybrid frameworks are critically evaluated. Benchmark performance acrossdatasets such as WESAD, DREAMER, DEAP, and FER2013 is compared using accuracy,precision, recall, and F1-score. Key challenges—including dataset scarcity, generalizability, realtime deployment, and ethical considerations—are discussed. Future directions highlight multimodalfusion, lightweight models for classroom integration, and privacy-preserving frameworks. Thiswork provides a structured foundation for advancing deep learning–based stress detection and fostering adaptive,emotionally intelligent educational environments.