MULTI-OUTPUT RANDOM FORESTS
Henrik Linusson · Borås Academic Digital Archive (University of Borås) · 2013
The Random Forests ensemble predictor has proven to be well-suited for solving a multitude of different prediction problems. In this thesis, we propose an extension to the Random Forest framework that allows Random Forests to be constructed for multi-output decision problems with arbitrary combinations of classification and regression responses, with the goal of increasing predictive performance for such multi-output problems. We show that our method for combining decision tasks within the same decision tree reduces prediction error for most tasks compared to single-output decision trees based on the same node impurity metrics, and provide a comparison of different methods for combining such metrics.