Experiments for a better Gini index splitting criterion for Data Mining Decision Trees algorithms

The paper attempts to find splitting criteria derived from the Gini index criterion that would improve the performance of the Gini index criterion. The splitting criteria are used in the Data Mining Decision Tree (DT) training process. Thus, in the process of training of DT on seven databases we used the Gini index (G) criterion and two criteria derived from it. On these databases, for each of the three splitting criteria, three kinds of DT were executed: unpruned, error-based pruned and pessimistic pruned. We considered the classification error rate and the number of decision rules corresponding to each of the nine types of DT built and executed on the seven databases. Following the experiments, it turned out that in the case of DT induced by the symmetric Gini index (SG) splitting criterion the decision rules number and the accuracy of the classification on the test records are better than in the case of the G splitting criterion and modified Gini index (MG) splitting criterion.