k-prototypes 군집분석에서 가중치 선택에 관한 연구

Abstract

The cluster analysis is an exploratory data analysis process of finding the characteristics of each group by dividing the objects constituting the data into several clusters having similar characteristics. This method maximizes homogeneity within a cluster and maximizes heterogeneity among different clusters. The criterion for selecting the cluster analysis algorithm depends on the attributes of the objects that constitute the data. If the object attributes of the data are a mixture of continuous and categorical types, then k-prototypes algorithm which is hybrid version of k-means algorithm that deals with continuous attributes and k-modes algorithm that deals with categorical attributes can be used. For the use of the k-prototypes clustering method, the weight for the distances calculated in the categorical property should be determined. In this paper, we propose a method to find the weight using the bootstrap method based on stability and investigate the applicability of the proposed method through simulation.