Predicting Attendance at Major League Soccer Matches: A Comparison of Four Techniques

Abstract

Sport team managers need to predict attendance levels at sporting events to plan staffing levels, plan inventories, and decide upon possible promotions. This paper discusses predicting attendance at Major League Soccer events using data from the 2014 and 2015 seasons. Panel data is obtained for each team, season, and weather category. A traditional least squared dummy variable linear regression technique is used along with three machine learning algorithms-random forest, M5 prime, and extreme gradient boosting. Extreme gradient boosting provides superior results with respect to out-of-sample root mean square error statistics. Well-founded technique for working with different methods is presented and the efficacy of contemporary algorithms is offered.