The model we used is a gradient-boosting, multi-output regression model. The underlying algorithm is based on many simple decision trees, combined in order to increase the accuracy of the predictions.