Using Collaborative Filtering Algorithms for Predicting Student Performance

Models that accurately predict student performance can be useful tools for planning educational interventions aimed at improving the results of the teaching-learning process, contributing to saving government resources and educators' and students' time and effort. In this paper it is studied the performance of collaborative filtering (CF) algorithms when applied to the task of student performance prediction. CF algorithms have been extensively and successfully used in recommender systems, but not in the considered educational scenario. The performances of two baseline methods and six state-of-the-art CF are compared when predicting if students would hit or miss multiple-choice questions, using two large educational datasets, created from the interaction between students and educational software. It was verified that CF algorithms account for consistently higher performance than the baselines for most metrics. Among the CF algorithms, memory based methods presented an overall better accuracy, precision, and recall. Nevertheless, all CF algorithms presented relatively low recall in identifying incorrect answers.