E-inclusion or digital inclusion is the European Commission's effort to ensure that everybody can benefit from the opportunities provided by digital technologies. Blended learning is a convenient and affordable way to learn new digital skills. However, the problem is the low number of e-learning graduates and the even lower number of those who use the newly acquired digital skills for professional purposes. To impact blended learning course results and to improve the usage of newly acquired digital skills machine learning and data mining techniques are used. The purpose of the study is to develop and compare e-inclusion predictions models, taking into account the following conditions: 1) the prediction model must use a set of data obtained from different digital skills blended training courses; 2) the prediction model must identify as many risk learners as possible at the same time to prevent the model from targeting actual non-risk learners as at-risk. We compared seven e-inclusion prediction models for blended learning courses. The models predict digitally excluded learners using classification, clustering, linear regression algorithms, and combinations of these algorithms. We used two metrics – F1 and F2 measures for prediction models. We concluded if the aim is predicting as many students at risk as possible then the classifier ensemble method with majority voting combined with linear regression predictions is appropriate. This model predicts 98.40% of digital excluded learners, however only 58.20% of those predicted as digitally excluded are actually excluded. Combining the classifier ensemble method with the majority voting and clustering model can predict as digitally excluded 82.80% of excluded students and can predict correctly 79.60% of learners as risk learners.