Development of Multiple Class Human-Computer Interaction System using Machine Learning Algorithm for Eyeball Movement

Abstract

Modern technologies in the field of Biomedical Engineering are flourishing astonishingly in recent times. Human-computer interaction (HCI) is one of the newest additions in this field. It is the study of the way people interact with computers and how the computers are or are not developed for interacting with human successfully. Electrooculography (EOG) machine can be used as HCI device. It is a technique for measuring the corneo-retinal standing potential which is present between the front and the back of the human eye. Pairs of electrodes are generally attached either above and below the eye or to the left and right of the eye to detect the eye movement. A potential difference occurs between the electrodes. Considering that the resting potential is constant, the recorded potential is a measure of the eye's position. EOG device can pick up these resting potentials while moving the eyeball in different directions. Data classification is important for HCI systems. It is to identify a new observation which belongs to a set of categories. Classification is done based on a training dataset having observations whose category membership is familiar. Various algorithms can be used to classify bio-signal data like ECG, EEG and EOG. These algorithms are called machine learning algorithm. It is a set of mathematical approaches to teaching computers to train based on large amount of data without step by step human instruction. In this research, EOG data are classified using machine learning algorithm to develop a multiple class HCI system. EOG data for different directional eyeball movement is acquired with the help of Biopac MP3X Acquisition unit. By placing the disposable surface electrodes on the right position of the skull and connecting all the leads and wires to the proper channel, the setup is ready to pick up the EOG data. Subjects are instructed to follow the LED sequence in the navigational setup. The data of 7 subjects aged between 22 to 48 years are taken for this experiment. The data is then saved using Biopac Student Lab Software and then preprocessed to prepare an EOG dataset for classification. With Weka 3.9.2, the classification procedure is done on the prepared dataset. Six classification algorithms i.e. naïve bayes, support vector machine, logistic regression, k-nearest neighbor, random forest and bagging are applied on the dataset. Comparison is shown among the algorithms based on different parameters. In the EOG dataset, features are also added which can be correlated with the classes. This correlation method is performed in IBM SPSS Statistics 25 software to find the most significant features related to the class. From the classification result, the accuracy of the different classifiers are obtained. The accuracy of Naïve Bayes is 30.7692%, SVM is 30.7692%, Logistic regression is 53.8462%, KNN is 7.6923%, Random forest is 84.61% and Bagging is 92.31% respectively. From the comparison among the classifiers based on different parameters, bagging has the highest and KNN has the lowest accuracy among them. The proposed method is then compared with other researches where it is seen that other methods applied only two or three algorithms but in the proposed method six machine learning algorithms are used. It is observed that bagging is more suiTable algorithm for EOG data than other algorithms used in the mentioned related works. As for the correlation, only the chi-square test is performed as Fisher’s exact test can be performed for 2x2 matrix whereas there are 9 classes in the dataset. From the chi-square test result, it is seen that the mean of channel 1 (horizontal channel) and channel 2 (vertical channel) used to acquire EOG data for eyeball movement are the most significant features. These two featuress are directly related to the classes.