Enhancing Everyday Seizure Detection: A Channel Reduction Approach

Epilepsy is a chronic neurological disorder characterized by unprovoked seizures. Developing seizure detection systems has become a focus, as they can aid in preventing accidents that may occur when epileptics lose consciousness during a seizure. Therefore, there is a need for these systems to operate in real time and be designed to enable patients to maintain their daily activities. Such advancements hold the potential to significantly enhance the quality of life for individuals affected by epilepsy. To that end, improving computational efficiency through channel reduction is critical. This study introduces an approach for selecting channels while preserving the accuracy of seizure detection. The seizure detection phase encompasses feature extraction and classification. Extracted features include the frequency domain based on Empirical Mode Decomposition (EMD). Machine learning classifiers such as Random Forest, SVM, and kNN were employed, and 10-fold cross-validation to training and testing to mitigate the overfitting problem. Optimal efficiency was achieved when employing three channels, as a result of channel selection, with each of the three classifiers. Utilizing three channels, the proposed model achieved performance, attaining 99.97% sensitivity, 99.98% specificity, and 99.99% accuracy. Notably, there were overlapping channels contributing significantly to the high performance observed. Leveraging this insight, this study proposes a selection of pivotal channels crucial for seizure detection, hence could advance the development of devices capable of detecting seizures in everyday life.