Splitting Data in Decision Trees Using the New False-Positives Criterion

Abstract

Classification is a widely used technique in various fields, including data mining and statistical data analysis. Decision trees are one of the most frequently occurring knowledge representation schemes used in classification algorithms. Decision trees can offer a more practical way of capturing knowledge than coding rules in more conventional languages. Decision trees are generally constructed by means of a top down growth procedure, which starts from the root node and greedily chooses a split of the data that maximizes some cost function. The order, in which attributes are chosen, according to the cost function, determines how efficient the decision tree is. Gain, Gain ratio, Gini and Twoing are some of the most famous splitting criteria used in calculating the cost function. In this paper, we propose a new splitting criterion, namely the False-Positives criterion. The key idea behind the False-Positives criterion is to consider the instances having the most frequent class value, with respect to a certain attribute value, as true-positives and all the instances having the rest class values, with respect to that attribute value, as false positives. We present extensive empirical tests, which demonstrate the efficiency of the proposed criterion.