Lightweight Seizure Detection for Wearable Electronics via Riemannian t-SNE

Accurate epileptic seizure detection is critical for emerging consumer healthcare devices. This work presents a lightweight framework based on phase space reconstruction (PSR) of EEG signals. Covariance matrices are mapped to a Riemannian tangent space for robust feature extraction. To address high dimensionality, a supervised t-distributed stochastic neighbor Embedding (t-SNE) strategy is utilized to maximize the Fisher ratio, significantly enhancing feature separability prior to k-nearest neighbor (KNN) classification. Experimental results demonstrate that the proposed method achieves superior accuracy with low computational complexity, making it highly suitable for real-time, resource-constrained consumer electronic applications.