Collaborative filtering is the method of making predictions of a user’s preferences based on passively collected data. Data used to train machine learning models to perform collaborative filtering are inherently sparse and models must be adapted to handle the data sparsity. In this study, we implement four models to perform collaborative filtering to predict the movies a user would like to watch. The predictions were made using user’s prediction of other movies as the only features. We analyze the models by the utility of its output for a recommendation system, the quality of the predictions, and the tractability of both training and prediction. Our results reinforce that the industry standard, neighborhood based collaborative filtering, is the best model to perform movie predictions.