A combinative method for decision tree construction

The selection of the splitting attribute in decision tree construction process is the key point for the size and quality of the tree. Although several criteria have been proposed and there are good papers that compare their results, no consensus have been adopted regarding the best method. In this paper we present a new approach in which each candidate attribute is evaluated using a set of available criteria and the attribute voted the best by most of the criteria will be selected as the winning splitting attribute. Each criterion will be evaluated based on the contingency table created at each splitting node. An approach based on OLAP is used for faster contingency table aggregation.