Evaluating skin condition using a new decision tree induction algorithm

Decision tree induction is well suited for applications requiring simple, explicit and intuitive classification structure. Due to the deteriorating generalization performance with increasing size and depth of the tree, construction of decision trees of small size and depth is a fundamental to widespread realization of the many benefits of decision tree based classification. In this paper we present a decision tree induction method based on a novel classifiability measure. The proposed algorithm makes a decision at a node based on the number of correctly classified instances at the node as well as the classifiability of the incorrectly classified instances. We demonstrate the efficacy of the proposed algorithm using a biomedical dataset in which optical images of human infant skin, coupled with localized noninvasive biophysical measurement of epidermal skin barrier properties are used to evaluate the health of the skin