Research on Catering Business Demand Forecasting Model Based on the Weather Sensitivity Theory and GBDT Algorithm

Abstract

In this paper, GBDT machine learning is introduced into the demand forecasting of catering e-commerce, and a weather-sensitive demand forecasting model based on GBDT is constructed. In addition, twenty-two influence factors are selected in the model, including minimum daily temperature, average temperature, highest daily temperature, minimum daily pressure, average pressure, highest daily pressure, 20-20 hours precipitation, hours of sunshine, extreme wind speed, direction of extreme wind speed, average wind speed, average aqueous vapor pressure, average relative humidity, maximum wind speed, direction of maximum wind speed, minimum relative humidity, logarithm of sales volume of last week, average difference in temperature, average difference in precipitation, promotion factor and holiday factor. Then, the model is applied to an empirical study to analyze the data and forecast the future demand. This paper also compares the results with several models including linear regression model, passenger flow model and BP neural network model. Finally, the experimental results show that the demand forecasting model based on GBDT has the highest accuracy and is an effective solution to solve the high demand fluctuation of catering e-commerce business.