Personalized Feature Selection for Wearable EEG Monitoring Platform

Electroencephalography (EEG) signal monitoring can be applied for many purposes, such as epileptic seizure detection. To design a reliable, wearable EEG monitoring platform for seizure detection in daily use, this paper presents a two-step approach to select a small subset of discriminative features from a few number of channels. In the first step, linear discriminant analysis (LDA) is applied to choose informative channels which have highly-ranked LDA criterion values. Then in the second step, the least absolute shrinkage and selection operator (LASSO) method is adopted to incrementally add features into selection subset. To determine the best number of channels and features for each subject, a personalization technique is utilized by evaluating the classification result of different feature subsets based on support vector machine (SVM) classifier. Experimentation on CHB-MIT database shows that on average, the proposed method selects approximately 3 channels and 7 features, and yields F-1 score of 81% based on SVM evaluation.