Influence of Feature Selection Methods on Classification Sensitivity Based on the Example of A Study of Polish Voivodship Tourist Attractiveness

Abstract

Abstract The purpose of this article is to determine the influence of various methods of selection of diagnostic features on the sensitivity of classification. Three options of feature selection are presented: a parametric feature selection method with a sum (option I), a median of the correlation coefficients matrix column elements (option II) and the method of a reversed matrix (option III). Efficiency of the groupings was verified by the indicators of homogeneity, heterogeneity and the correctness of grouping. In the assessment of group efficiency the approach with the Weber median was used. The undertaken problem was illustrated with a research into the tourist attractiveness of voivodships in Poland in 2011.