Vladimir Nikulin – Associate Professor, Department of Mathematical Methods in Economics, Vyatka State University.
Address: 36, Moskovskaya syr., Kirov, 610000, Russian Federation.
E-mail:  vnikulin.uq@gmail.com

Svetlana Palesheva – Student, Department of Mathematical Methods in Economics, Vyatka State University.
Address: 36, Moskovskaya syr., Kirov, 610000, Russian Federation.
E-mail:  s.palesheva@gmail.com

Dar’ya Zubareva – Student, Department of Mathematical Methods in Economics, Vyatka State University.
Address: 36, Moskovskaya syr., Kirov, 610000, Russian Federation.
E-mail:  ZubarevaDasha@yandex.ru

Relevance of the problem is caused by the necessity to determine a suitable direction for the students (attendees of training courses), who are practically "living" in the Internet, by taking into account the growing number and availability of research and educational resources. The paper considers an algorithm that has been tested in real-time during the international data analysis competition VideoLectures.Net ECML / PKDD 2011 (Track 2) on the TunedIT. VideoLectures.Net platform. It is an open multimedia resource of video lectures with research and educational content. Open social and educational systems provide new opportunities for millions of students to enjoy high quality educational materials in real-time. There are a lot of lectures read by outstanding scientists and researchers within the framework of the most significant and well-known events such as conferences, summer schools, workshops and scientific activities in the various scientific fields. The main objective of the Internet resource is to promote scientific ideas and knowledge exchange, which is achieved by providing high quality educational materials not only to a scientific community, but also to a wider audience. All lectures, documents, information and links are systematized and grouped by editors, taking into account users' comments.

We suggest using two lectures (out of three of the group) in order to determine the direction of the forecast, which includes a set of predicted lectures, supplemented with appropriate frequency or empirical probability. Accordance of the forecasted set is calculated so as to predict the remaining third lecture. Further improvements have been achieved through the application of uniform ensembles based on random subsets (this approach is also known as bagging).