Novel hybrid feature selection algorithms for diagnosing erythemato-squamous diseases

Abstract

This paper proposes hybrid feature selection algorithms to build the efficient diagnostic models based on a new accuracy criterion, generalized F-score (GF) and SVM. The hybrid algorithms adopt Sequential Forward Search (SFS), and Sequential Forward Floating Search (SFFS), and Sequential Backward Floating Search (SBFS), respectively, with SVM to accomplish hybrid feature selection with the new accuracy criterion to guide the procedure. We call them as modified GFSFS, GFSFFS and GFSBFS, respectively. These hybrid methods combine the advantages of filters and wrappers to select the optimal feature subset from the original feature set to build the efficient classifiers. To get the best and statistically meaningful classifiers, we not only conduct 10-fold cross validation experiments on training subset, but also on the whole erythemato-squamous diseases datasets. Experimental results show that our proposed hybrid methods construct efficient diagnosis classifiers with high average accuracy when compared with traditional algorithms. © 2012 Springer-Verlag.