Probably the core of this paper is the usage of “correlation-based feature selection (CFS) plus a k-nearest neighbors (KNN) data mining algorithm.” The authors compare it with some other machine learning algorithms for feature selection, such as naive Bayesian, and then show and conclude that CFS+KNN is the most accurate and fastest of them all, and not subject to the curse of dimensionality (unlike the naive Bayesian method). The experiments are described, including how the electroencephalography (EEG) signals were obtained and then exported from MATLAB to Weka through a script named the Matlab2Weka toolbox.

So, this paper is important both for cognition sciences, showing how to detect signals of attention during the learning process, and also for computer scientists, because it is a good reference for showing the comparison of different classification and feature extraction methods. For these reasons, I highly recommend reading it.