Developing fuzzy classifiers to predict the chance of occurrence of adult psychoses

One of the key objectives behind implementation of computer logic in medicine is to mimic doctors' medical logic. The present work is a novel attempt to develop fuzzy logic-based expert systems (ESs). which are able to reason like doctors for screening adult psychosis. Among several techniques of fuzzy classifier (FC)-design, clustering-to-classifier technique (CCT) has been adopted, in this paper. We have clustered a set of statistically generated psychosis data (with 24 factors and 7 responses) using (i) fuzzy C-means (FCM) algorithm, (ii) entropy-based fuzzy clustering (EFC) algorithm and its proposed extensions. The properties of the best set of clustered data are then utilized to develop the respective FCs. The number of rules of the FC is made equal to the number of clusters obtained above and the attributes of the cluster centers carry information of the rule base of the said FC. Moreover, a genetic algorithm (GA) has been used to tune the data base of the FC for further improvement of its performance. The performances of these FCs are tested on a set of randomly-gene rated test psychosis cases and another set of diagnosed cases. It is found that for both the data sets, each of the FCs is appreciably accurate in inferring and the classifier developed based on FCM-clustered data slightly outperforms the FC developed from EFC-clustered data. It may happen due to the fact that the performance of the developed FC depends on the nature of clusters also. (c) 2008 Elsevier B.V. All rights reserved.