Building an optimal individual educational trajectory for students is currently impossible without the use of recommender systems. The role of such a system for individual educational trajectory is to inform the student about the most useful disciplines for him, based on his interests and academic performance. An important factor for recommendations is the forecast of student progress. Through assessments, the student's inclination to a particular subject, as well as the level of his education, can be expressed. In addition, the success of mastering the discipline greatly affects the motivation of students. Thus, an important part of the recommender system is the prediction model. The article describes an approach to developing a system for predicting the progress of university students in elective disciplines. The predictive model is based on machine learning algorithms. Collaborative filtering was used as the main method. The sources for collecting data on the digital footprint of students are the electronic educational information environment of the university and the official portal, which hosts educational standards, work programs and annotations. Information about students and their progress is presented in the form of three tables – a rating plan, disciplines and a list of students. The data structure has the form of a double nested dictionary, where the keys are sections of the university from faculty to specialty and year of study, and the values are tables with the student identification number, course of study, subject, normalized grade, and elective and model labels. To solve the filtering problem, k-means algorithms, cosine proximity, and Pearson correlation were used. The applied approaches are able to work for small data and do not require large performance costs. The resulting predictive model has a sufficiently high accuracy and can be used in recommender systems to build individual educational trajectories of university students.