Power-Efficient Epileptic Seizure Detection Using Linear Predictive Coding for Wearable Applications

Epilepsy is a critical neurological disorder affecting millions worldwide, requiring accurate and timely detection to prevent life-threatening complications. Clinical devices achieve high accuracy in seizure detection, ensuring reliable medical monitoring. However, the demand for wearable devices necessitates lightweight, low-power, real-time solutions. Wearable EEG-based seizure detection requires efficient signal encoding to optimize power consumption while maintaining classification accuracy and computational efficiency. In this study, we hypothesize that applying Linear Predictive Coding over long EEG segments provides a computationally efficient approach suitable for wearable applications. To evaluate this, EEG signals were analyzed using Linear Predictive Coding, Discrete Wavelet Transform, and Power Spectral Density-based features, and classified using Multilayer Perceptron, Random Forest, and Support Vector Machines. Among the tested combinations, the Linear Predictive Coding and Random Forest model achieved the best energy efficiency with an average consumption of 2.73 microjoules per percent and classification accuracy of 93.18%. One-way analysis of variance showed no significant accuracy difference among feature extraction methods (p = 0.856) but revealed a significant difference in energy efficiency (p = 1.93 × 10⁻⁷⁵). These findings demonstrate that Linear Predictive Coding is a promising technique for wearable seizure detection, offering a balance between accuracy and energy efficiency for next-generation medical applications.