Clustering quality criterion based on the features extraction of a tagged sample with an application in the field of brain-computer interface development

In applied machine learning, the problem of sample heterogeneity is often encountered. For example, the sample heterogeneity leads to serious difficulties in solving the problem of brain electrical activity patterns recognition when developing a brain-computer interface for people of different social characteristics.
In this work, we proposed a new criterion of clustering quality based on the features selection, which has low computing needs and is based not on the proximity / remoteness of the sampled objects, but on the ability of an algorithm to recognize hidden patterns, that is, to select groups that are similar in features. We have shown the areas of practical application of the algorithm, in particular, in the task of brain electrical activity patterns recognition when pronouncing 8 words by people with different social characteristics.