Customer segmentation in insurance industry using modularity-based mapper clustering algorithm

Abstract

Clustering is an unsupervised learning method to divide the whole data set into subgroups with specific patterns using a distance between data points. In recent years, the insurance industry has observed the high utilization of their customer data sets as they find the analytics, such as customer profiles, based on clustering can be profitable. However, an insurance data set contains very private information and the de-identification is required. Then, the shape of a data set becomes complex. Although there are some existing clustering methods applied to insurance data sets, any dominant method is yet to be seen because of the size and complex shape of this data set. Along this line of research, we propose the modularity-based mapper clustering algorithm. In our method, we focus on the structure of a data set. We generate the structure of a data set using a mapper algorithm, one type of TDA, and analyze the structure with modularity, the concept of a network analysis. Experimental results show our clustering has a better performance than other clustering algorithms in terms of the association rule mining, RFM analysis, cluster validity index, and distribution of important features. This study shows the potential of combining the TDA method and network analysis theory.