An Investigation of Different Machine Learning Approaches for Epileptic Seizure Detection

Wearable devices increasing popularity provide convenient alternatives to healthcare services outside hospital premises. Wearables provide enhancements for automatic tools to assist physicians during patient diagnosis, treatment, and many other situations with limited costs and computing resources. In this context, in-device processing using machine learning algorithms can accelerate syndromes monitoring such as epilepsy detection and minimize risks of privacy disclosure due to extended data transmission to cloud servers. In this paper, we investigate the performance of five machine learning algorithms, i.e., Support Vector Machine (SVM), Random Forest (RF), Naive Bayes (NB), K-Nearest Neighbor (KNN), and Neural Network (NN), in terms of accuracy to diagnose a syndrome and the computational cost to embed it in a wearable device. We tested the algorithms in the classification of an Electroencephalography (EEG) sampled dataset available at the UCI machine learning repository. From the results, we concluded that SVM and RF have good accuracy in identifying epileptic seizures from the EEG dataset. Additionally, only RF fulfills the low computational cost required to embed such applications in-device.