Evaluation of the Feature Space of an Erythematosquamous Dataset Using Rough Sets

Abstract

The di®erential diagnosis of erythematosquamous diseases remains a di±cult task requiring both clinical and histopathological data to support a diagnosis. The principle reason for diagnostic ambiguity is based on the signi¯cant degree of overlap in the overt symptoms of this class of disease. Histopathological evidence can assist in making a positive diagnosis - but is labor and resource intensive. In order to evaluate the diagnostic veracity of clinical versus histopathological features of erythematosquamous diseases, a comparison of both fea- tures classes was evaluated using rough sets. The results indicate that the histopathological feature space provided a much more signi¯cant classi¯cation rate relative to clinical features. In addition, the results of this preliminary study indicate that only a small subset of the histopathological feature space is required for maximal classi¯cation accuracy. Key words and phrases. Datamining, Dermatology, Erythematosquamous Diseases, Reducts, Rough Sets.