An Optimized Quadratic Support Vector Machine for EEG Based Brain Computer Interface

Abstract

The Brain Computer Interface (BCI) has a great impact on mankind. Many researchers have been trying to employ different classifiers to figure out the human brain's thoughts accurately. In order to overcome the poor performance of a single classifier, some researchers used a combined classifier. Others delete redundant information in some channels before applying the classifier as they thought it might reduce the accuracy of the classifier. BCI helps clinicians to learn more about brain problems and disabilities such as stroke to use in recovery. The main objective of this paper is to propose an optimized High-Performance Support Vector Machines (SVM) based classifier (HPSVM-BCI) using the SelectKBest (SKB). In the proposed HPSVM-BCI, the SKB algorithm is used to select the features of the BCI competition III Dataset IVa subjects. Then, to classify the prepared data from the previous phase, SVM with Quadratic kernel (QSVM) were used in the second phase. As well as enhancing the mean accuracy of the dataset, HPSVM-BCI reduces the computational cost and computational time. A major objective of this research is to improve the classification of the BCI dataset. Furthermore, decreased feature count translates to fewer electrodes, a factor that reduces the risk to the human brain. Comparative studies have been conducted with recent models using the same dataset. The results obtained from the study show that HPSVM-BCI has the highest average accuracy, with 99.24% for each subject with 40 channels only.