Research On Prediction Of Fresh Food Sales Based On ARIMA-LightGBM-NGBoost Hybrid Model

With the change of time,we have entered the era of big data and information.In the era of big data,the one who has the information has the world.If companies can make reasonable use of information to predict future trends and make corresponding countermeasures,they will be able to increase their profits.Fresh produce,as a typical instant commodity,is perishable and difficult to store,so it is particularly important to forecast sales and manage inventory.However,most fresh food supermarkets in China are still using conventional methods to forecast sales,and some companies are relying on the subjective judgement of their managers,resulting in the perishability rate of fresh food in China being much higher than that of developed countries.The sales of fresh food are often affected by a variety of factors,so conventional methods and approaches are no longer able to meet the needs of fresh food forecasting.In the above context,this study focuses on two issues: the factors influencing fresh food sales forecasting and fresh food sales forecasting models.This study develops a research on the influence of weekend holiday and weather factors on fresh food sales forecasting,and constructs a hybrid ARIMA-Light GBM-NGBoost forecasting model considering weekend holiday and weather factors.This study confirmed that weekend holiday and weather factors can improve the forecast accuracy of fresh food sales,and that weather factors have different effects on the improvement of forecast accuracy on weekdays and weekends,with a slower improvement of forecast accuracy on weekends,but a faster improvement of forecast accuracy on weekdays.On the basis of the above,this study established a hybrid ARIMA-Light GBM-NGBoost forecasting model,which has better applicability and higher forecasting accuracy,and the model has a smoother error rate,less dependence on the sample and greater ability to generalise.At the same time,it also exhibits strong robustness in its ability to smoothly analyse long and short term sales data,and its benefits arguably become more pronounced as the length of the training data expands.In a sense,the study extends the theory and methods related to fresh food sales forecasting and lays the foundation for future in-depth research.The results of this study will also provide a theoretical and practical basis for the development and implementation of production and management activities in the fresh food industry in China.