Interpretable behavioral clusters of gamblers through unsupervised learning.

Area of Science:

• Computational Social Science
• Behavioral Economics
• Data Science in Gambling Research

Background:

• Understanding gambling behavior heterogeneity is key for effective harm reduction.
• Previous research often treats high-involvement gamblers as a monolithic group.
• Electronic Gambling Machines (EGMs) represent a significant area of gambling research.

Purpose of the Study:

• To segment high-intensity EGM users into distinct behavioral subtypes using unsupervised machine learning.
• To identify key behavioral indicators that differentiate these subtypes.
• To inform personalized responsible gambling initiatives through data-driven insights.

Main Methods:

• Applied Uniform Manifold Approximation and Projection (UMAP) for dimensionality reduction.
• Utilized Density-Based Spatial Clustering of Applications with Noise (DBSCAN) for cluster identification.
• Employed random forest classifiers to determine discriminative features for each cluster.

Main Results:

• Identified four distinct behavioral clusters among high-intensity EGM users.
• Clusters characterized by: impulsive/night-time play, consistent high-frequency play, structured high-stakes play, and rapid binge-like activity.
• Key discriminative features included balance trajectory, inter-session timing, and transaction interval variability.

Conclusions:

• High gambling involvement is heterogeneous, comprising diverse behavioral subtypes.
• This segmentation provides a framework for personalized responsible gambling interventions.
• Findings support data-driven player protection strategies in the gambling industry.