Informative trees by visual pruning

The aim of this study is to provide visual pruning and decision tree selection for classification and regression trees. Specifically, we introduce an unedited tree graph to be made informative for recursive tree data partitioning. A decision tree is visually selected through a dendrogram-like procedure or through automatic tree-size selection. Our proposal is a one-step procedure whereby the most predictive paths are visualized. This method appears to be useful in all real world cases where tree-path interpretation is crucial. Experimental evaluations using real world data sets are presented. The performance was very similar to Classification and Regression Trees (CART) benchmarking methodology, showing that our method is a valid alternative to the well-known method of cost-complexity pruning.