Identification and Detection of Real-Time Behavioral Shifts in Gambling, Through Hidden Markov Models

Sammanfattning

Online gambling has increased in popularity, raising the urgency for safeguards that can flag risky play as it unfolds rather than after months of data have accumulated. This work proposes a real-time approach based on Hidden Markov Models (HMMs) that infer hidden “behavior states” from minute-level slot-machine logs. Two model families are trained and compared, varying both the number of hidden states and the mix of input features. Results show that a ten-state specification captures virtually all behavioral variation observed within a single session. The most informative predictors are, the three-minute standard scores of mean bets, bet variance and deposit amount, the number of bets placed, elapsed session time and sales channel. Calendar variables such as day of the week add negligible explanatory power. The resulting models not only predict the next state but more importantly, identify recent state sequences that reveal states of interest for example, high-risk bursts of play. These findings demonstrate the feasibility of live behavior tracking and lay the groundwork for future real-time, player-specific harm-minimization interventions.