Concept analysis with granular computing method

Medical Decision Support Systems (MDSS) are intelligent software systems that are able to make deduction in case of lack of information and in the presence of ambiguity. In order to model the uncertainty in these systems, different soft computing methods like Bayesian networks, rough sets, neural networks, fuzzy logic, inductive logic programming, genetic algorithms and/or hybrid systems that are combination of several of these mentioned methods are used. Bayesian network is a data oriented method that is frequently used in MDSS.In this study, based on the similarity between Bayesian network and concept lattice in formal concept analysis, a data oriented model reflecting the relations between Symptom-Disease is proposed, in which diseases and symptoms are modeled as objects and attributes respectively.In order to test the developed model, with the technical problems of finding data sets, an ALARM network based on real situations and real patient data is used. Data sets of various size is implemented using NETICA software. In order to reduce computational complexity, ROSETTA software, which is a fundamental tool in rough set theory and based on discernibility, is used.Lattice of symptom-disease concept is constructed using related algorithms by Oracle JDeveloper software and later the conditional probabilities that are indicator of the models accuracy are calculated by Bayes theorem. Correct diagnosis is achieved when the actual situation in the network and the result of the program is the same. The proposed model applied to various sized data set can give the correct diagnosis with an average of 51% accuracy.Also, these datasets are applied to the machine learning methods like C4.5, Support Vector Machines (SVM) and Multi Layer Perceptron (MLP). Support Vector Machine has given the best result as approximately 75% average accuracy.