C-3DP: A cross-cluster analysis model to identify latent categorical customer attributes

Abstract

Using two-dimensional clustering methods to segment a customer database is a popular practice. The advantage of two-dimensional clustering is the ability to map customers according to a well-defined business logic. For example, how many segments can be identified based on the customers' age group and RFM score? Such an approach also has the advantage of reducing the dimensionality of datasets and a model's training time. Conversely, the trade-off of clustering on two dimensions is to ignore all the other dimensions potentially available in a CRM or web analytics platform. As such, the qualitative traits analysis of each segment from available customer dimensions can be challenging, especially for categorical dimensions with higher cardinality. In order to maximise the customer insights derived from cluster analysis, the paper proposes a qualitative trait prevalence scoring system: the C-3DP index (categorical density, dominance and diversity prevalence index). This technique maps a subset of dominant qualitative segment traits using a simple algorithm, as opposed to relying solely on traditional descriptive analytics approaches.