비선형 패턴을 지닌 범주형 자료의 군집분석을 위한 그래프 기반 거리 측도

Abstract

Purpose: The purpose of this study is to suggest a more efficient distance measure taking into account the patterns of data for clustering categorical data Methods: The proposed categorical geodesic distance is calculated with three main steps: (1) The first step measures the Gower distance between two observations composed of categorical variables. (2) The second step is to represent the data as a mutual k-nearest neighbor graph. (3) The final step calculates the distance between two observations with the shortest path in the graph. The distance obtained from (3) is utilized for clustering categorical data. In particular, our proposed method is suitable for data with nonlinear patterns. Results: Our experimental results using several real-life datasets reveal that the categorical data also has implicit topological structures and confirm that the distance considering implicit data patterns generally yields better clustering performance than existing Gower distance measure. Conclusion: This study revealed that the adoption of the data patterns using our proposed distance measure positively affected the results of cluster analysis.