Using an Alternative Coefficient of Concordance for Features Clustering

Abstract

Clustering can play an important role to assess the consistency of expert opinion or features. It allows to identify consistent groups and the value of the relationship between them. The analysis of the obtained clusters allows drawing certain conclusions about the dependencies of features and making adjustments to the expert survey. A hierarchical divisive algorithm can be used for feature clustering. In this case, Kendall's coefficient of concordance can be used as an indicator of the closeness of the features relationship. The simplicity of its calculation contributes to its wide application in various fields of research. The disadvantage of this coefficient includes the fact that it shows a deviation from cases of complete inconsistency which can't always be considered the presence of a good consistency. The application of an alternative coefficient of concordance to the clustering of signs is considered. It takes into account precisely the deviation from the case of complete consistency. Different clustering comparison obtained using the Kendall's coefficient of concordance and the alternative coefficient of concordance is made using the written software implementation of the divisive hierarchical method for some indicators from the “Science and Innovations” section of the socio-economic indicators of Russian regions for 2020. Based on this comparison it is concluded that the alternative coefficient may produce better clustering than the Kendall's coefficient of concordance.