By looking at Figure \ref{984894}, it is possible to observe that the labeled data is separated to some degree. Therefore, to make it easier for the classification process a machine learning model can be used to help in such a task. The selection of the classification model is presented in the next section. 

Machine Learning Algorithm Selection for EOG classification 

Machine learning is a subfield of artificial intelligence that has two important branches, supervised and unsupervised learning. Supervised learning is preoccupied with problems that have a known or specific response based on a determined set of observations or inputs, while unsupervised learning tries to find patterns or clusters in unlabeled data, in other words, it is used to draw inferences from datasets. Supervised learning is categorized depending on the type of problem that it is intended to be solved. If the problem posses continuous output based on a specific set of inputs, the problem is classified as a regression problem. One of the most common examples of machine learning models for regression problems is linear regression. More generally, regression machine learning techniques are used to create a model to solve quantitative problems\cite{Faul_2019,Rebala_2019}
On the other hand, if the problem has a discrete output based on a specific set of inputs it is described as a classification problem. An example of a classification algorithm is a logistic regression that can be viewed as the counterpart of linear regression but instead of generating a linear model, the input and output relationship is map into sigmoid function.  In general, a classification model serves to solve quantitative problems. For the case of the eye-tracking device, the problem of monitoring eye behavior can be seen as a classification problem, which based on a set of features it is expected to know the movements of the users' eyes.  To find the best classification machine learning model, different types of algorithms were tested. These methods are briefly described and applied in the next section\cite{Faul_2019,Rebala_2019}.  To apply de algorithms MATLAB was used as the developing and testing environment. 

K- nearest neighbor  (K-NN)

The K nearest neighbor is a classification technique in which it is assumed that similar things are close together. The concept of closeness is referred to as the distance between two points (Euclidian distance). If the distance between the new observation point is closest to one of the classes of the data set, it is assumed that the new data belongs to that class.  In this case, with the sampled data and features, K-NN algorithm was tested for the classification task with 3 nearest neighbors.K-fold cross-validation with k equal to 5, was used as the resampling method to avoid overfitting problems.  Overfitting refers to the modeling error that happens when a resultant function is fit very closely to a limited set of data points or observations. \cite{Rebala_2019}
In this case, four metrics were calculated based on the Confusion Matrix, these were  Accuracy, Precision, Recall, and F1 score. The obtained metrics can be appreciated in Table \ref{604641}