Research On Sales Data Mining And Forecasting Model Based On Deep Learning

Sales forecasting and decision analysis are very important to the company.Reasonable sales forecasting can affect the production,purchase,distribution,inventory and other links of the enterprise,and save a lot of money for the enterprise.With the development of enterprise management informationization,sales data and influencing factors of sales are stored in the enterprise internal management system.Full use of these data can not only predict sales,but also make sales guidance.Selling data is essentially time series data,and the sequence model in deep learning can capture the time lapse.This paper improves Trend Alignment with Dual-Attention Multi-Task Recurrent Neural Networks,redefines the sales forecasting problem,and adds the known information of the forecasting time point,such as whether the time point is a holiday or not.In this paper,the input of decoder phase is obtained by splicing the known information of predicted time point and the result of attention mechanism.The experimental data are predicted by using the commodity sales data of a large supermarket and the pesticide sales data of a pharmaceutical enterprise.The experimental results show that the accuracy of the improved algorithm is improved compared with the original model.And from the prediction results of supermarket,with the increase of the length of decoding stage,the value of the improved algorithm increases compared with the original algorithm.In order to facilitate enterprises to predict and analyze sales data,sales data prediction and analysis decision-making system named FOREUNDERSTANDING is developed.Enterprise managers can upload the company's sales data and use the system.The system is divided into four functional modules as follows: data visualization module,feature analysis module,prediction module and sand table maneuver module.The design of sand table maneuver module is to save the model in the training stage.Drag and modify the original data,such as promoting the product,increasing the number of sales activities,etc.After modification,call the model to get the final prediction results and observe the impact of the change,which can assist the sales staff to make analysis decisions.