Generating Better Decision Trees

A new decision tree learning algorithm called IDX is described. More general than existing algorithms, IDX addresses issues of decision tree quality largely overlooked in the artificial intelligence and machine learning literature. Decision tree size, error rate, and expected classification cost are just a few of the quality measures it can exploit. Furthermore, decision trees of varying quality can be induced simply by adjusting the complexity of the algorithm. Quality should be addressed during decision tree construction since retrospective pruning of the tree, or of a derived rule set, may be unable to compensate for inferior splitting decisions. The complexity of the algorithm, the basis for the heuristic it embodies, and the results of three different sets of experiments are described.