Clothing Sales Forecast Considering Weather Information

Under the background of global warming and frequent extreme weather events,this paper proposes to establish a clothing sales forecast model considering weather information under the policy route that China will promote the development of "carbon peaking" and lead the green development of the clothing industry.In general,the impact of weather on clothing sales shows a high degree of irregularity,and scholars generally use traditional statistical models or improved linear models,using comprehensive data,and monthly or quarterly data.Some scholars also use artificial intelligence and other forecasting methods,but most of them only incorporate weather as a covariate into the model,do not focus on the relationship between weather variables and sales forecasts,and do not further quantify the impact of weather on sales forecasts of different clothing categories.Therefore,this article will improve on this basis,focusing on the clothing industry,especially the impact of weather information on clothing sales forecasts.This paper uses day-level data,adds body temperature and wind chill index data based on basic weather data,and marks holiday data to construct a more complete and comprehensive characteristic index system.For the total sales of clothing,basic clothing,and seasonal clothing,the "rolling time window" is used to divide the data set,and multiple machines learning single models are trained at the same time to quantify the impact of weather on the sales prediction of different clothing.According to the experimental results,it is found that adding weather information in the three machine-learning single models of random forest,XGB and GBDT does not always improve the accuracy of model sales prediction.For basic clothing,adding weather information will increase the model error,and the model MSE will increase by 1.53% on average;For seasonal clothing,adding weather information can reduce model errors,and model MSE is reduced by an average of 10.21%.Secondly,according to the prediction results of the single model,an ensemble learning model fused with a stacking strategy is established to improve the prediction effect of the single model.In the case of filtered weather information in the feature,compared with the single model,the Stacking model reduced the prediction error MSE of basic clothing sales by 12.74%,the forecast error of seasonal clothing sales decreased by 17.85%,and the forecast error increased with the increase of the forecast time.Because seasonal clothing is affected by weather and local sales fluctuate greatly,to solve this problem,we introduce the Attention mechanism to build a neural network model,highlight the impact of important feature data on clothing sales trading days,improve the model’s prediction of seasonal clothing sales anomalies,and further improve the accuracy of sales forecasting.The clothing sales prediction model considering weather information can help sales people actively respond to market changes brought by weather and adjust store operation strategies;It can provide production planning guidance for workshop production supervisors to help enterprises better plan production costs.It can help enterprises prevent overproduction caused by demand fluctuations caused by climate change,alleviate inventory backlog from the source,improve the economic efficiency of enterprises,and respond to the call of the country for green and low-carbon.