University of Amsterdam publishes free open-source algorithm for KSA to estimate online casino player risk

Researchers from the University of Amsterdam have released a free open-source algorithm that estimates risky behaviour in online casino players from actual gameplay data. For regulators and PSPs watching duty-of-care exposure, the interesting part is not the machine learning buzzword; it is that the Dutch regulator, Kansspelautoriteit (KSA), plans to use the tool as an independent benchmark against operators’ own risk models.

1. The model looks at player behaviour rather than declarations or questionnaires. It analyses betting patterns, including how much and how often someone bets; frequency and timing, such as playing at night for days in a row; and loss and win streaks, plus how players react to them. Based on those patterns, it calculates a risk score.
2. All forms of online gambling are included in the model. The KSA intends to use the risk scores to compare them with the scores used by providers, which matters because it gives the regulator a second set of eyes on whether an operator’s internal monitoring is doing the job.
3. The researchers say the tool is free and open source, unlike commercial providers of similar systems. The model, code and methodology have been made publicly accessible on the KSA website, so other parties can use them without relying on the closed systems of casinos themselves.
4. The researchers said that since the legalisation of online gambling in the Netherlands, the market has grown explosively. They also cited recent figures from the French regulator showing that around 60 per cent of online casino revenue comes from excessive gamblers, defined here as people who gamble frequently and for extended periods, with major financial, psychological and social consequences.
5. The tool was developed by PhD candidate Charles de Leau with UvA professors Reinout Wiers, from Psychology, and Johan Bollen, from Computer Science. De Leau said his own experience of gambling addiction’s impact on family and friends inspired him to pitch the concept to ZonMw, which funded the project from the KSA’s Addiction Prevention Fund.

The source says the model was trained on a “massive amount of data” obtained under Dutch gambling legislation, but does not give the final training set details. For operators, the practical takeaway is straightforward: if a regulator has an open benchmark for risky-play detection, “our vendor says so” becomes a weaker defence than it used to be.