Visual Classification: An Interactive Approach to Decision Tree Construction

Satisfying the basic requirements of accuracy and understandability of a classifier, decision tree classifiers have become very popular. Instead of constructing the decision tree by a sophisticated algorithm, we introduce a fully interactive method based on a multidimensional visualization technique and appropriate interaction capabilities. Thus, domain knowledge of an expert can be profitably included in the tree construction phase. Furthermore, after the interactive construction of a decision tree, the user has a much deeper understanding of the data than just knowing the decision tree generated by an arbitrary algorithm. The interactive approach also overcomes the limitation of most decision trees which are fixed to binary splits for numeric attributes and which do not allow to backtrack in the tree construction phase. Our performance evaluation with several well-known datasets demonstrates that even users with no a priori knowledge of the data construct a decision tree with an acc...

• Type: Conference
• Year: 1999
• Where: KDD
• Authors: Mihael Ankerst, Christian Elsen, Martin Ester, Hans-Peter Kriegel
• Keywords: Data Mining, Decision Tree, Decision Tree Classifiers, KDD 1999, Tree Construction Phase