Early detection of problem gambling based on behavioral changes using shapelets

Abstract

Recent years have seen strides achieved in the field of behavior analysis by using online gambling data. However, studies on time-series behavioral changes remain inadequate. In this study, we propose a classifier that quantifies changes in the player’s time series of online gambling behavioral data by using distance measurement with shapelet for the early detection of behaviors in players that could lead to problem gambling. We investigated the prediction capabilities of shapelets that represent behavioral change patterns, and the results showed that shapelet features can improve predictive accuracy. Furthermore, based on this result, we found characteristic behavioral changes leading to problem gambling, such as loss chasing. Subsequently, we demonstrated a possibility for improvements in accuracy using these behavioral change patterns based on expert knowledge.