Filter Based Feature Selection for Automatic Detection of Erythemato-squamous Diseases

This paper presents an automatic diagnosis model of erythemato-squamous diseases. The proposed model consists of two stages. In the first stage, two filter based feature selection methods, namely rough set using Johnson's algorithm and ranked features for feature selection of erythemato-squamous diseases are employed to select the optimal feature subset from the original feature set for dimensionality reduction in order to further improve the diagnostic accuracy. Next, for the sake of comparison, the diagnoses decisions are made by four different classification algorithms: k-nearest neighbors, Naive Bayesian classifier, linear discriminant analysis and decision tree. Experimental results show that the accuracies of the four base classifiers using ranked features outperformed those using rough set with Johnson's algorithm and the base classifiers without using feature selection. Using erythemato-squamous diseases dataset taken from UCI (University of California at Irvine) machine learning database. The accuracies of these four classifiers using ranked features on test sets (50% of the dataset) are 97.21, 98.32, 96.09, and 98.32, respectively. Therefore, we can conclude that the ranked features method is very promising in detection of erythemato-squamous diseases compared to the rough set using Johnson's algorithm and also compared favorably with previously reported results. This tool enables doctors to differentiate six types of erythemato-squamous diseases using clinical and histopathological parameters obtained from a patient.