Evaluating the generalizability of payment behavioral profiles across gambling brands

Abstract

ABSTRACT A gambler’s payment behavior – the deposit and withdrawal of funds – precedes and follows the act of gambling. Given this separation, the methods and results of machine learning models built at the payment-level could be better generalized across gambling formats. With this study, we empirically evaluated this notion by validating a cluster analysis across two independent datasets of digital wallet payment transaction records. Using a discovery dataset comprising 2,286 customers of a casino-focused Internet gambling brand, the k-means algorithm revealed five distinct payment profiles. Using a validation dataset comprising 5,580 customers of a sports-focused Internet gambling brand, we evaluated the generalizability of the discovery payment profiles. Specifically, we assessed validity by (1) clustering the validation dataset using the discovery method, (2) classifying the validation dataset into the discovery clusters, and (3) assessing the stability of cluster membership. Two large low risk clusters were validated across datasets. Three smaller potential risk clusters were only partially validated. Our findings suggest that gamblers’ payment behaviors are somewhat representative of their gambling behavior and may reflect dynamics of certain gambling formats. Stakeholders employing data science methods across gambling populations should be mindful of specific contexts and tailor analyses accordingly.