Machine Learning Prediction of Pokemon Go Exergame Enjoyment

Physical inactivity is on the rise, with dire consequences such as obesity and many diseases. Exergames, which try to gamify physical activity, have been demonstrated to be an effective solution. For exergames to be effectively improve health, they have to be played for extended periods of time. However, prior work has found that while many players enjoy exergames initially, they often suddenly stop playing. For instance, nearly 85 % of exergame players quit after a day, and 95% within 3 months. We believe that if timepoints at which a player stops enjoying an exergame can be predicted from objective data, mitigating action such as recommending a new exergame, can be taken to sustain the user engagement. In this paper, we explored whether player enjoyment of Pokemon Go can be predicted using machine learning analyses on location, step count, game usage and session statistics and weather data. We found that Bagged trees performed best overall, predicting player Exergame Enjoyment Questionnaire (EEQ) scores with a $\mathbf{7 4 \%}$ accuracy.