Adaptation of a cluster discovery technique to a decision support system

Abstract

This paper reports on the implementation of a cluster discovery technique to a decision support system. The model is a multi-criteria multi-alternative decision environment. The decision support reported here is a diagnostic system. Adaptive Resonance Theory 1 (ART1), which is a cluster discovery neural network, was employed to cluster related symptoms to identify a specific disorder. ART1 was then modified to determine the degree of the belongingness of a new set of values to a cluster. For the purpose of demonstrating the functionality of the system, a set of symptoms for various diseases of the eye was utilized to create different clusters. Thus, each cluster represents a sub-type of an eye disorder. The closeness of a set of values representing a user's symptoms to a particular cluster is first determined and a rank ordering of the degree of membership to each cluster is returned. The cluster with the highest rank is reported as the disorder. The proposed clustering approach avoids major problems faced by the traditional multicriterion decision making practices. First, the clustering approach is immune to the problem of interdependence of the criteria. Second, the clustering approach avoids the problems arising from the criteria being measured along different dimensions. Finally the clustering approach is unaffected by heterogeneity of the criteria. The system is implemented as a web application and the functionality of its implementation in the teaching environment as well as use by lay persons is discussed.