Industry Research Brief Vol 2 (1) — Behavioural Feature importance

The research note describes the behavioural markers most consistently predictive of harm across BetBuddy machine learning models.

Categories

The research explores which behavioural patterns are most predictive of gambling-related harm, based on machine learning models developed by BetBuddy and deployed in live gambling environments. Key findings:

• Top Predictors of Risk: The most influential behavioural markers across five models include:

• Frequency and amount of deposits
• Frequency of play (days played, number of bets)
• Increases in deposit amounts
• Use of multiple payment types
• Session time and bet size
• Declined deposits (e.g. failed payment attempts)

• Diverse Feature Categories Matter: The top 16 features span deposits, game selection, responsible gambling tool usage, and session metrics. Deposit-related and play frequency features were the most consistently important.

• Trends and Volatility Are Key: Models performed best when they included not just current behaviour levels but also trends (e.g. increasing deposits) and variability (e.g. inconsistent play patterns).

• Model Performance: The models achieved high accuracy (average test set AUROC ~95%) and were tailored to specific operators and player bases.

• Responsible Gambling Tool Usage: Limit-setting changes were important in some models, but other RG tool usage features were generally less predictive.

Implications for industry and policy:

• Focus on deposit and play frequency data when identifying at-risk players.

• Use behavioural insights to personalize player feedback and interventions.

• Develop models tailored to specific products and player segments.

• Incorporate financial transaction data and trend/volatility metrics for better risk detection.

• Avoid over-reliance on single features; a diverse set of behavioural indicators improves model accuracy.

• Continue evaluating and refining models with independent research and across diverse player bases.

These insights support the development of more effective, personalized responsible gambling strategies using behavioural data and machine learning. For full details, please refer to the complete document.