Cognitive ability is a very complex neurological process and at the same time very studied, especially in the process of attention due to the high impact it has on the quality of life in people, however, the detection and classification of attention levels remain a challenge due to the scarcity of data to work with, as well as the subjectivity of the indicators when measuring the levels of attention. The present work intends to develop an Artificial Intelligence model applied to eye trajectory data for the classification of attention levels. We worked with a public database, using a Convolutional Neural Network to extract the features of the trajectories, using the feature vectors as input data in different Machine Learning models; in addition, a methodology was proposed to generate and validate the increase of data due to the imbalance of classes in the database. The logistic regression model was selected to perform the classification due to its adaptability to the data after testing, in addition to being optimized using the Grid Search technique. In conclusion, a hybrid model was developed and optimized incorporating Deep Learning techniques with Machine Learning, proving that a model can accurately classify reaching an accuracy higher than 95%, surpassing previous research. This study demonstrated the potential of working with non-invasive techniques using Artificial Intelligence in the classification of attention levels.