Improvement of decision tree generation by using instance-based learning and clustering method

A new classifier, which can be regarded as a modification of an existing top-down decision tree generation approach C4.5, is proposed. It utilizes a clustering method as a preprocessing and a k-Nearest Neighbor rule as a complementary classifier to C4.5 applied to each cluster. Experiments on several standard data sets demonstrate improvements of performance of the new classifier compared with that of C4.5.