Dealing with decision costs in CBR in medical applications

Abstract

This paper presents an approach for developing a Web-based decision support system that aids general practitioners in diagnosing dermatological problems using case-based reasoning. Our research aims to address the lack of decision support in current store-and-forward telemedicine systems and the need to deal with different costs of decisions. We propose a cost calculation method for minimizing the total misclassification cost. Two different techniques of measuring cost are evaluated. A series of tests is carried out on different sets of cost matrices for comparing the two methods. Our results show taking max values gives a lower misclassification cost and still manages to achieve a slightly higher accuracy compared to the summing method of instance-weighting.