Time-series Feature Extraction by Shapelets and Prediction of Problem Behavior in Online Gambling

In recent years, the field of behavior analysis using online gambling data has developed. However, researches on time-series behavioral changes are inadequate. In this study, we propose a classifier that quantifies the changes of the player’s time series of online gambling behavioral data using distance measurement with Shapelet for the purpose of early detection of players leading to problem gambling. Especially, we investigate the prediction capabilities of local Shapelets which represent short term user behavior characteristics and global Shapelets representing long term ones. Prediction experiments show that the time series features were more effective than the non-time series features, and also show that the case using both the local time series features and the global time series features performed the best.