Acquisition and Comparison of Classification Algorithms in Electrooculogram Signals

In recent years there has been considerable advances in the development of assistive technologies to enable the Person with Disabilities (PwD) better participate in society and interact. One way to allow PwD interaction is using interfaces based on the electrical potentials that our body presents, called biopotentials. The acquisition and the classification of these are crucial steps in the development of these technologies. In this work, a setup to acquire electrooculogram (EOG) signals was developed to track eye movements for a minimally invasive computer/game interface. This work includes setup and creation of a dataset, feature extraction of the collected signals and the implementation of classification algorithms. Three machine learning algorithms were designed to classify the EOG signals based on its characteristics: Softmax Regression, Gaussian Discriminant Analysis (GDA) and K-Nearest Neighbors (KNN). Comparing the algorithms classification performances, KNN presented the best results with 88.9% overall accuracy.