Granular Computing Combined with Support Vector Machines for Diagnosing Erythemato-Squamous Diseases

A computational model with a new hybrid feature selection approach is developed in this paper to determine the type of erythemato-squamous disease. The new feature selection method combines the strength of granular computing (GrC) and support vector machines (SVM) together with the advantages of filters and wrappers to select the optimal feature subset to build a sound classifier. We treat the erythemato-squamous disease dataset as a decision information system, where the sample features are considered as condition attributes and the class label the decision attribute. We calculate the granular of each feature and decision attribute, then evaluate the significance of each feature to classification by the difference between its granularity and that of decision attribute, after that we rank features in descending order by their significance. Generalized sequential forward search (GSFS) strategy together with SVM is adopted to select the necessary features to condense decision information system without compromising its classification capacity. 5-fold cross validation experiments have been conducted on the erythemato-squamous disease dataset taken from UCI (University of California Irvine) machine learning repository. Experimental results demonstrate that our diagnostic model has got condensed decision information system for erythemato-squamous disease with less features than the original ones while achieving a comparable accuracy in the literature.