Analysis and Comparison of Classification Methods in a Brain Machine Interface
In this chapter, the analysis and comparison of four classification methods for a simplified Brain-Machine Interface (BMI) are presented. The BMI involves the use of only one electroencephalography (EEG) bipolar connection: O1- P3. The EEG signal is processed to extract features of the alpha wave (7.19 14.4Hz) in order to determine whether the subjects have their eyes closed or open. The signal processing technique is based on a Discrete Wavelet Transform (DWT) algorithm that allows processing the data on-line while the EEG signal is being recorded. The features that are extracted using DWT are then classified in order to estimate the subjects’ eyes state. The four classifiers here compared are: Naive Bayes (NB), Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), and Adaptive Neuro-Fuzzy Inference System (ANFIS). A set of 2029 input vectors were considered for the comparison of the classifiers, they were divided as follows: 1771 for training (60%), 609 for testing (30%) and 203 for validation (10%). The best classification accuracy of the validation data was obtained by the NB classifier (93.59%) and the worst classification accuracy by the ANFIS classifier (88.17%). However, using a contingency table it was shown that the SVM obtained the best performance to classify eyes closed, which was consider the most important feature to classify in the BMI application used. The final output of the classifiers is the input of a microcontroller that generates a pulsewidth modulation signal that opens or closes a robotic hand if the subjects’ eyes are open or closed, respectively.
Unable to display preview. Download preview PDF.
- Chang, S., Yu, H.: Novel probability neural network. IEEE Proceedings – Vision, Image and Signal Processing 152(5), 137–144 (2005)Google Scholar
- Hsu, W.Y.: Neuro-fuzzy prediction for brain-computer interface applications. In: Fuzzy Inference System-Theory and Applications, pp. 299–312 (2012)Google Scholar
- Jahankhani, P., Revett, K., Kodogiannis, V.: A rule based approach to classification of eeg datasets: a comparison between anfis and rough sets. In: 9th Symposium on Neural Network Applications in Electrical Engineering, pp. 157–160 (2008)Google Scholar
- Lopez-Arce Vivas, E., Garcia-Gonzalez, A., Galindo, F., Sanchez-Gonzalez, V.: Algorithm to detect six basic commands by the analysis of electroencephalographic and electrooculographic signals. In: Proc. IEEE Conference on Biomedical Robotics and Biomechatronics, Rome, Italy, pp. 189–194 (2012)Google Scholar
- López-Arce, E., García-González, A., Figueroa, I., Fuentes, R.: Discrete wavelet transform and anfis classifier for brain-machine interface based on EEG. In: Proc. IEEE Conference on Human System Interaction, Sopot, Poland, pp. 137–144 (2013)Google Scholar
- Semmlow, J.: Biosignal and medical image processing. CRC Press, EE. UU (2008)Google Scholar