How can predictive methods for forecasting product categories of the next purchases of online customers of a telecommunication company be combined into ensembles in an accuracy-enhancing way? If this topic sounds interesting to you, we would like to refer you to the just published research article "Hybrid Recommender Systems for Next Purchase Prediction Based on Optimal Combination Weights", co-authored by Nicolas Haubner and Thomas Setzer.
Combining Logistic Regression, k-Nearest Neighbor classification, Neural Networks, CART Decision Trees, Random Forests, Adaboost as well as Gradient Boosting, weighting schemes are learned from real world data that minimize the Brier score of probability estimation for products of different categories and increase the accuracy over the individual models.