Performance Improved EEG based vowel recognition using Integrated PCA and Quadratic SVM

Abstract

In this paper, performance improved EEG based vowel recognition using integrated Principal Component Analysis (PCA) with Quadratic Support Vector Machine (SVM) is developed with the help of a Graphical User Interface (GUI). The GUI displays Time-Frequency plot which shows suppression of alpha, beta and gamma rhythms. There is also a plot showing the Event Related Potential (ERP) of the patient. In addition to that, the GUI produces a topographical plot which shows the power of the electrical signals at the electrode. Data around the electrode is extrapolated. In the proposed system, data is collected by EEG device at sampling rate of 128Hz. This new sampling rate will produce lower samples for each recording, thus computational power and time required to train and test classifiers is reduced drastically. Further, five classifiers tested are built based on One Versus Rest (OVR) strategy, which are Fine Decision Tree (FDT), Linear Discriminant Analysis (LDA), Quadratic Support Vector Machine (QSVM), Weighted kernel Neural Network (WkNN) and also Subspace Discriminant Classifier (SDC). The performance for these classifiers are evaluated in recognizing the vowels imagined: “a, e, i, o , u”. QSVM is the best classifier among the five and shows that PCA has proven to improve the quality of classifier with 90.1% accuracy on 10 trials of 10 subjects tested. This improvement is significant as it boosts the performance for approximately 20% in accuracy. The system also allows specialist to monitor their patient’s brain activity which is recorded by the EEG device.