Title:e-Fold Cross-Validation for Recommender-System Evaluation

Abstract:To combat the rising energy consumption of recommender systems we implement a novel alternative for k-fold cross validation. This alternative, named e-fold cross validation, aims to minimize the number of folds to achieve a reduction in power usage while keeping the reliability and robustness of the test results high. We tested our method on 5 recommender system algorithms across 6 datasets and compared it with 10-fold cross validation. On average e-fold cross validation only needed 41.5% of the energy that 10-fold cross validation would need, while it's results only differed by 1.81%. We conclude that e-fold cross validation is a promising approach that has the potential to be an energy efficient but still reliable alternative to k-fold cross validation.