Graphical Representation and Exploratory Visualization for Decision Trees in the KDD Process

Abstract

This article presents a proposal of graphical representation and exploratory visualization scheme for Decision Trees (DT) in the KDD (Knowledge Discovery in Database) process, specifically in the data mining stage. This pursues the improvement of the understandability of the internal model operation, which is a key issue in pattern exploration data mining tasks. This exploratory visualization is based on the well-known technique named Tree map (maps of trees), that allows representing hierarchical structures like DTs, in which grids are used to represent their node structure. The proposed visualization represents the number of instances or weight associated to a node with a color scale in degradation, using either 2D o 3D views of the data-mining model. In this sense, this work main contribution is to introduce a new visualization scheme that allow humans a better understandability of how the data mining models work internally.