Demand forecasting model development through big data analysis

Abstract

This study aims to predict the visit rate of game customers in order to efficiently deploy dealers to for table games in foreign-only casinos. The goal is to predict the number of customers visiting each casino game event by date and business hour, and ultimately, to predict the number of game tables necessary per game. Once the number of game tables has been decided upon, the number of game dealers going into the game will be decided, thus enabling efficient dealer operation. The number of games at the casino in question is nine, and the hours of operation are divided into three sections per day. The data used in the analysis are data on customer game history from January 2013 to December 2015 in certain areas of foreigner-only casinos in Korea, and external environment data such as holidays per country, exchange rate information, and weather conditions were utilized as well. In addition, the correlation between variables was analyzed and the key derived variables were utilized. As an analytical tool, SAS's Big Data analysis tool was used. The prediction shows that some games have a 10-16.2% prediction error based on MAPE. Further, by MAE, there are no more than five absolute errors. Further research could result in more accurate predictions if data such as customer's gaming preferences, environmental variables of casino businesses, and national situations were added to the analysis.