SPECIAL EXCLUSIVE DRAFT RESOLUTION OF THE MECHANISM FOR THE DETECTION OF RISK BEHAVIORS (III)
SURVEILLANCE: Four monthly phases and the XGBoost algorithm
The technical core of the Draft Resolution submitted for processing by the Directorate General for the Regulation of Gambling reveals with absolute precision the operational architecture that must be executed by all authorized operators.
The mechanism will not operate statically or through annual reviews, but on a monthly and recurring basis within the first seven calendar days of each month. In each execution, the companies' computer systems must analyse a retrospective time window of 183 continuous days on each and every one of the active user accounts, structuring the screening through four consecutive phases of mathematical discrimination.
The first phase functions as an objective exclusion filter for the so-called sporadic gamblers. Those users who in the 183-day window concurrently meet three minimum conditions are exempt from complex analytical evaluations and classified ex officio as risk-free profiles: having added less than two days of effective activity on the platform, having participated in less than five sessions within the casino or slot games offer and registering less than ten sports or horse racing bets. Once this first sieve has been passed, the second phase applies a direct discriminant algorithm for the immediate detection of critical situations: if the cut-off equation calculated on the financial variables F2 (net balance of profit and loss), F4 (global deposited volume) and F12 (frequency of successive top-ups with residual balance) satisfies the mathematical relationship ?0.00276 · F2 ? 0.0002477 · F4 + 0.3951199 · F12 ? 5.484472, the participant immediately acquires the status of risk player without requiring further treatment.
For the remaining mass of regular players, the third phase requires the appraised extraction of more than 60 objective variables and pattern indicators, and the fourth phase culminates with the execution of the predictive model based on enhanced decision trees (XGBoost - Extreme Gradient Boosting), calibrated and supplied directly by the DGOJ. The algorithm computes speed ratios in top-ups, occasions when the wallet falls below 2 euros followed by immediate deposits, recurrence in requests for increase in limits in the face of reductions, peaks in intensity in specific time slots such as the one between 8:00 and 8:59 hours and the segmentation of the product through binary variables that differentiate whether 70% of the expenditure is channeled towards sports betting or towards virtual slot and roulette terminals, issuing a final probability that determines in an automated and unappealable way the level of risk of the user.
