Enhanced nearest centroid model tree classifier

Abstract In this study, first, we improved an existing variant of the Nearest Centroid algorithm. In this new version, the predictive power of features and within-class variances are used as weights in distance calculation. This version is called the Enhanced Nearest Centroid (ENC). Second, we proposed a new model tree algorithm for binary classification. It is named as the Enhanced Nearest Centroid Model Tree (ENCMT). The model tree is built using ENC at each leaf node of the decision tree. To evaluate the performance of the new model tree, we used an independent test platform and ran the algorithm on 30 binary datasets available therein. Results showed that ENCMT improves the performance of the decision tree algorithm. We also compared ENCMT with the Logistic Model Tree (LMT) algorithm and showed that it outperforms LMT as well. We also designed a bagging algorithm where ENCMT is used to build a random forest. Our comparison results show that its performance is significantly better than the Random Forest (RF) algorithm.