PurposeThis paper aims to deal with the previously unknown prediction accuracy of students’ activity pattern in a blended learning environment.Design/methodology/approachTo extract the most relevant activity feature subset, different feature-selection methods were applied. For different cardinality subsets, classification models were used in the comparison.FindingsExperimental evaluation oppose the hypothesis that feature vector dimensionality reduction leads to prediction accuracy increasing.Research limitations/implicationsImproving prediction accuracy in a described learning environment was based on applying synthetic minority oversampling technique, which had affected results on correlation-based feature-selection method.Originality/valueThe major contribution of the research is the proposed methodology for selecting the optimal low-cardinal subset of students’ activities and significant prediction accuracy improvement in a blended learning environment.