This study aims to produce a more personalized recommendation system for online learning using multicriteria in collaborative filtering and data from the Binus Online Learning repository as a knowledge base. The study uses forecasting (regression) and consists of three stages: (1) collecting data on the results of the learning process; (2) adding context data; and (3) providing recommendations based on decisions generated from a rule. Recommendations are made based on mapping the distribution of online learning material topics and the assessment rubrics contained in the course outline in each course. This study uses a comparative evaluation of several measurement errors using six regression models, namely (a) a generalized linear model, (b) deep learning, (c) a decision tree, (d) random forests, (e) gradient-boosted trees, and (f) a support vector machine. The results show that the most suitable model that provides the best prediction is the gradient-boosted trees model. Prediction results on student grades are used to provide recommendations that contain links to material that can be learned by students.