DETECTING GAMBLING IMPULSIVITY USING A BILSTM AUTOENCODER

Session Title

AI in Gambling: Mapping the Research Landscape & Advancing Behavioral Risk Detection

Presentation Type

Paper Presentation

Start Date

26-5-2026 12:00 AM

Abstract

Traditional behavioral risk models often rely on surface-level classification, failing to capture the complex temporal dynamics of gambling addiction. This study explores time-series analysis to uncover hidden behavioral phenotypes within a dataset of 2,286 U.S. online gamblers. I propose a Bidirectional Long Short-Term Memory (BiLSTM) Autoencoder for unsupervised anomaly detection. By engineering sequential features, specifically measuring ‘manic velocity’, loss of control, and financial distress, the model learns to reconstruct ‘normal play patterns’, flagging deviations as impulsive anomalies. Quantitative results demonstrate that the BiLSTM identifies a high-risk segment (top 5%, n = 118) characterized by rapid churning and behaviors that traditional baselines, such as Isolation Forest and K-Means, fail to capture. These findings validate that mathematically encoding the temporal dimension of gambling behavior provides a framework for identifying and preventing high-risk gambling habits.

Author Bios

Graduate Research Assistant, UNLV International Gaming Institute

Share

COinS
 
May 26th, 12:00 AM

DETECTING GAMBLING IMPULSIVITY USING A BILSTM AUTOENCODER

Traditional behavioral risk models often rely on surface-level classification, failing to capture the complex temporal dynamics of gambling addiction. This study explores time-series analysis to uncover hidden behavioral phenotypes within a dataset of 2,286 U.S. online gamblers. I propose a Bidirectional Long Short-Term Memory (BiLSTM) Autoencoder for unsupervised anomaly detection. By engineering sequential features, specifically measuring ‘manic velocity’, loss of control, and financial distress, the model learns to reconstruct ‘normal play patterns’, flagging deviations as impulsive anomalies. Quantitative results demonstrate that the BiLSTM identifies a high-risk segment (top 5%, n = 118) characterized by rapid churning and behaviors that traditional baselines, such as Isolation Forest and K-Means, fail to capture. These findings validate that mathematically encoding the temporal dimension of gambling behavior provides a framework for identifying and preventing high-risk gambling habits.