Gambling Insider: BetBuddy and City University’s Research Using Predictive Algorithms in Gaming

The interview explains how machine learning is helping the gambling industry protect customers through accurate predictions.

Categories

The research presents a novel method for interpreting machine learning models used to predict gambling-related harm, specifically through the use of “feature risk curves.” These curves help explain how individual behavioural features (e.g. night-time play or declined deposits) influence the model’s predictions of serious self-exclusion risk, offering a more transparent and interpretable approach to understanding black-box algorithms in responsible gambling contexts. Key findings:

• Feature risk curves provide directional insights into how specific behaviours affect predicted risk, holding other variables constant.
• Night-time play is only weakly associated with increased risk up to a point (38–68% of play at night), after which risk declines—suggesting stable night play (e.g. shift workers) may not be harmful.
• Deposit decline ratio shows only a modest and non-significant increase in risk, challenging assumptions that frequent declined deposits are always strong harm indicators.
• The method enables domain experts to challenge and refine model assumptions, offering more nuanced interpretations than feature importance rankings alone.
• The approach is scalable, model-agnostic, and applicable to other domains beyond gambling, supporting explainable AI in sensitive applications.
• The analysis highlights that models based on self-exclusion may effectively identify some at-risk players but miss others, underscoring the need for diverse harm proxies.

These insights support the use of explainable AI tools like feature risk curves to improve transparency, trust, and effectiveness in gambling harm prediction models. For full details, please refer to the complete document.