Visualizing the random forest by 3D techniques

Abstract

Random forest which contains a set of decision trees is a popular method in data mining. It has the advantages of high accuracy, high learning speed and the ability of dealing with high dimensional data. The decision model from the training process, however, is non-deterministic because of the sampling process. Although we can calculate correlations between different decision trees to infer the performance, it's not comprehensive to non-specialists. So, the goal of this project is to find a way of visualizing the learning process and the final model using 3D techniques. As a consequence, it can help in model selection by visualizing the patterns of different trees in terms of density, similarity and so on. Moreover, it can help users to understand how rules are learnt and then applied in decision making. Finally, it can provide an interactive interface for manual modifications (e.g. pruning). © 2012 Springer-Verlag.