Improving SVM Classification on Imbalanced Datasets for EEG-Based Person Authentication

Abstract

Support Vector Machine (SVM) has been widely used in EEG-based person authentication. Current EEG datasets are often imbalanced due to the frequency of genuine clients and impostors, and this issue heavily impacts on the performance of EEG-based person authentication using SVM. In this paper, we propose a new bias method for SVM binary classification to improve the performance of the minority class in imbalanced datasets. Our experiments on EEG datasets and UCI datasets with the proposed method show promising results.