Classification of EMG Signals of Eye Movement Using Windowing Technique and Cubic SVM

In this paper an approach to classify electromyography of extraocular muscle signals for six eye movement classes, namely, Blink, Normal Behaviour, Left, Right, Downward and Upward Movement is discussed. The dataset contained two signal values, one by horizontally connected electrodes and another by vertically connected electrodes. The technique was applied on both the signal values individually but resulted with less accuracy for vertically connected electrodes. Windowing technique was applied to extract features from pre-processed data stream. Total 28 features were calculated from pre-processed dataset and formed a feature matrix with the class label. To prevent overfitting during training, the rows of the feature matrix were randomized. This work has been compared to existing works of literature and has been found better. The classification accuracies were evaluated by various classifier algorithm, but the best accuracies achieved were 96.8% by Cubic Support Vector Machine.