RAFT (RAndom Forest Tool) is a new java-based visualization tool designed by Adele Cutler and Leo Breiman for interpreting random forest analysis. RAFT uses the VisAD Java Component Library and ImageJ. Our work in developing RAFT was funded, in part, by NSF ITR 0112734.
The central plot in RAFT is a 3-dimensional scatterplot of the MDS coordinates obtained from the random forests proximity matrix. This scatterplot can be colored and brushed, and the brushed data points are highlighted in associated parallel coordinate displays and heatmaps. Brushing allows us to easily determine variable importance in specific regions of the data, and to look at input variables for subgroups of data that may be present in the scatterplot. Parallel coordinate displays are used to represent input variables and variable importance, and these displays can be enhanced by alpha-blending. Heatmaps are used to represent the votes for each class, and for interactions and proximity matrices (if available). Information from selected datapoints can be output to a file for future use, and the images can be saved as jpeg files. For a more detailed description, please see the screenshots. At the present time, RAFT is only available for random forest classifiers. We have plans to develop a version for regression forests in the coming months.
On Windows, download and install
Java 2 Runtime Environment (JRE).
Java3D Runtime for the JRE (select the OpenGL Runtime for the JRE).
Then download and run this installer, which does not bundle the JRE.
The source code for RAFT is freely available:
RAFT Source Code.
VisAD Source Code.
ImageJ Source Code.
Currently no FAQ.