Sensitivity analysis of multi-attribute decision making methods in Clinical Group Decision Support System

Abstract

A development of a clinical group decision support system (CGDSS) has been carried out for diagnosing both neurosis and personality disorders. The knowledge, stored in the knowledge base, were generated from the aggregated preferences given by decision makers. Two types of preferences used here, i.e. the preferences of a mental evidence by a mental condition; and the preferences of a mental disorder by mental condition. Ordered weighted averaging operator was adopted to aggregate those preferences. This aggregation process was carried out after transforming the selected subset to fuzzy preference relation format. Then the Bayesian theorem was adopted to compute the probability of evidence given a particular disorder. After developing the knowledge base, the next step is to develop an inference engine. The method used for developing an inference engine is multiattribute decision making concept, this is because of the system was directed to choose the best disorder when a particular condition was given. Many methods have been developed to solve MADM problem, however only the SAW, WP, and TOPSIS were appropriate to solve problem here. In this knowledge base, the relation between each disorder and evidence were represented X matrix (m x n) that consist of probability value. Where the X <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">ij</sub> was probability of jth mental evidence given i <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">th</sup> mental disorder; i=1,2,...,m; and j=1,2,...,n. Sensitivity analysis process was to compute the sensitivity degree of each attribute to the ranking outcome in each method. The sensitivity analysis was aimed to determine the degree of sensitivity of each attribute to the ranking outcome of each method. This degree implies that there were a relevant between an attribute and a ranking outcome. This relevant attribute can be emitted by influence degree of attribute C <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">j</sub> to ranking outcome f <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">j</sub> . Then, relation between sensitivity degree and influence degree for each attribute, can be found by computing the Pearsonpsilas correlation coefficient. The biggest correlation coefficient shows as the best result. This research shows that TOPSIS method always has the highest correlation coefficient, and it is getting higher if the change of the ranking is increased. The experimental results shows that that TOPSIS is the appropriate method for the clinical group decision support system for the above purposes.