Impurity Measurement in Selecting Decision Node Tree That Tolerate Noisy Cases
Abstract. In a recent years, recommending an appropriate attribute of binary decision tree under unusual circumstances – such as training or testing with noisy attribute, has become more challenge in researching. Since, most of traditional impurity measurements have never been tested how much they can tolerate with encountered noisy cases. Consequently, this paper studies and proposes an impurity measurement which can be used to evaluate the goodness of binary decision tree node split under noisy situation, accurately. In order to make sure that the accuracy of decision tree classification by using the proposed measurement has been yet preserved, setting up an experiment to compare with the traditional impurity measures was conducted. And the result shows that accuracy of the proposed measurement in classifying a class under noisy case is acceptable.