Forecasting Customers Visiting Using Machine Learning and Characteristics Analysis with Low Forecasting Accuracy Days

Abstract

In this paper, the number of customers visiting restaurants is forecasted using machine learning and statistical analysis. There are some researches on forecasting the number of customers visiting restaurants using past data on the number of visitors. In this research, in addition to the above data, external data such as weather data and events existing in ubiquitous was used for forecasting. Bayesian Linear Regression, Boosted Decision Tree Regression, Decision Forest Regression and Random Forest Regression are used for machine learning, Stepwise is used for statistical analysis. Among above five methods, the forecasting accuracy using Bayesian Linear Regression was the highest. The forecasting accuracy did not tend to improve even if the training data period was extended. Based on these forecasting results, the characteristics of days with low forecasting accuracy are analyzed. It was found that the human psychology around the payday and the reservation customers affected the number of visitors. On the other hand, the weather data such as temperature, precipitation and wind speed did not affect the accuracy.