Exploring new methods of harm analysis with Artificial Intelligence: Application of Recurrent Neural Networks, Clustering and Supervised Learning to classify behavioural events and anomalies in real-time
Session Title
Session 1-4-B: Technology and Harm Minimization
Presentation Type
Event
Location
Caesars Palace, Las Vegas, Nevada
Start Date
28-5-2019 3:30 PM
End Date
28-5-2019 4:55 PM
Disciplines
Artificial Intelligence and Robotics
Abstract
Traditional approaches in the application of machine learning to detect harm have used supervised machine learning techniques to train models using historic player data. These techniques typically prove most useful when a reasonable amount of historic player data is used to define model features. Whilst some of these models have proved effective at predicting events such as long-term self-exclusion, they often miss a large proportion of at-risk players who close their accounts or self-exclude soon after registration and who experience episodic bursts of intensive activity that can lead to harm.
In an effort to improve the state-of-the-art, BetBuddy and City, University of London, applied a variety of artificial intelligence techniques, including deep learning, to a UK online casino real money gambling data set to undertake i) streaming data analysis to predict events and/or flag anomalies, ii) session analysis as players approach self-exclusion, and iii) very early prediction following registration.
The research results have proved promising. Predicting self-exclusion using ‘day 1’ data, that includes a on site activity and demographic data, provided competitive model performance when benchmarked to PwC / GambleAware PGSI research. The application of recurrent neural networks was able to demonstrate the ability to predict real-time events and anomalies.
Keywords
online casino, recurrent neural network, supervised learning, artificial intelligence, problem gambling, anomaly detection
Funding Sources
The research was funded by Playtech Plc. The research design adn execution was overseen by BetBuddy (part of Playtech Plc) and City, University of London. The research was undertaken by three MSc student interns and overseen by Simo Dragicevic (BetBuddy) and Professor Garcez, Dr Weyde, and Dr Turkay (all City, University of London).
Competing Interests
Simo Dragicevic is employed by BetBuddy, a wholly owned subsidiary of Playtech Plc, a significant industry supplier and operator. Simo is also an External supervisor at City, University of London Current and previous research projects funded by Kindred Group, InnovateUK, EPSRC, ESRC and DSTL.
Exploring new methods of harm analysis with Artificial Intelligence: Application of Recurrent Neural Networks, Clustering and Supervised Learning to classify behavioural events and anomalies in real-time
Caesars Palace, Las Vegas, Nevada
Traditional approaches in the application of machine learning to detect harm have used supervised machine learning techniques to train models using historic player data. These techniques typically prove most useful when a reasonable amount of historic player data is used to define model features. Whilst some of these models have proved effective at predicting events such as long-term self-exclusion, they often miss a large proportion of at-risk players who close their accounts or self-exclude soon after registration and who experience episodic bursts of intensive activity that can lead to harm.
In an effort to improve the state-of-the-art, BetBuddy and City, University of London, applied a variety of artificial intelligence techniques, including deep learning, to a UK online casino real money gambling data set to undertake i) streaming data analysis to predict events and/or flag anomalies, ii) session analysis as players approach self-exclusion, and iii) very early prediction following registration.
The research results have proved promising. Predicting self-exclusion using ‘day 1’ data, that includes a on site activity and demographic data, provided competitive model performance when benchmarked to PwC / GambleAware PGSI research. The application of recurrent neural networks was able to demonstrate the ability to predict real-time events and anomalies.