Selecting latent factors when predicting student performance in online campus by using recommender systems

Recommender Systems is a popular method of Collaborative Filtering in Machine Learning. They are applied to many applications for prediction purposes, for example, the Predicting Student Performance problem. In this problem, we start from a database where the scores of some academic tasks (exercises, tests, exams) are stored for the corresponding students, considering a particular subject in the academic course. Since some scores for a particular student could be unknown (tasks not completed, submitted or evaluated), they can be predicted considering his behavior in the remaining tasks and the behavior of the other students for these unknown tasks. The prediction is performed by means of Matrix Factorization where the error (Root Mean Squared Error) is minimized by the Gradient Descent algorithm. Nevertheless, two problems arise: the best selection for the main parameters of the algorithms (learning rate and regularization factor) and the best selection for the number of latent factors. In this work, we propose a direct search of the optimal value of latent factors where the calculation for each number of latent factors is driven by a metaheuristic that select, at the same time, the optimal values of learning rate and regularization factor. Using this method, we can determine the best number of latent factors to be applied in further predictions for the similar databases.