Explainable Contrastive Learning Framework for EEG-Based Affective Computing in Privacy-Preserving Consumer Devices

Integration of emotion recognition into consumer electronics, such as smart watches and wearable EEG headsets, offers transformative potential for digital health through affect-aware applications in mental wellness, stress monitoring, and adaptive human–device interaction. However, existing approaches often lack interpretability, neglect privacy safeguards and are unsuitable for real-time deployment on resource-constrained platforms. This paper presents an explainable contrastive learning framework tailored for real-time emotion recognition on privacy-preserving consumer devices. Neurophysiologically grounded features, including differential entropy across canonical EEG bands and frontal asymmetry indices, are extracted from a reduced frontal-central montage optimized for wearable compatibility. A supervised contrastive learning module encodes latent representations that discriminate against emotions and are classified using a lightweight, low-latency predictor. Post-hoc interpretability is achieved via SHAP (SHapley Additive exPlanations), offering localized, instance-level feature attribution. Privacy preservation is ensured through a combination of pseudonymization, differential privacy during inference, and blockchain-based consent logging. Using the DEAP dataset, the framework achieves 91% accuracy in valence classification with sub-40 ms inference latency on edge-compatible hardware. This work delivers a unified architecture that integrates real-time emotion sensing, explainable machine learning, and privacy engineering, advancing the deployment of trustworthy affective computing in next-generation consumer electronics.