Analysis of Cluster Migrations Using Self-Organizing Maps

Discovering cluster changes in real-life data is important in many contexts, such as fraud detection and customer attrition analysis. Organizations can use such knowledge of change to adapt business strategies in response to changing circumstances. This paper is aimed at the visual exploration of migrations of cluster entities over time using Self-Organizing Maps. The contribution is a method for analyzing and visualizing entity migration between clusters in two or more snapshot datasets. Existing research on temporal clustering primarily focuses on either time-series clustering, clustering of sequences, or data stream clustering. There is a lack of work on clustering snapshot datasets collected at different points in time. This paper explores cluster changes between such snapshot data. Besides analyzing structural cluster changes, analysts often desire deeper insight into changes at the entity level, such as identifying which attributes changed most significantly in the members of a disappearing cluster. This paper presents a method to visualize migration paths and a framework to rank attributes based on the extent of change among selected entities. The method is evaluated using synthetic and real-life datasets, including data from the World Bank.