Induction of monotonic decision trees

The monotonie classification is a problem that widely exists in our real lives, e.g. venture capital, bank loans and credit assessment. As one of the classifiers, decision trees are easy to understand and implement. However, owing to the ignorance of the monotonie relationship between samples, traditional algorithms of decision tree training are not suitable for the monotonie classification. In this paper, we first discuss limitations of traditional decision trees for the monotonie classification, and then present a solution called the MGain. Experimental results show that the MGain not only generates a monotonie decision tree, but also yields a better performance in comparison to traditional decision tree algorithms for the monotonie classification.