Research On E-commerce Sales Prediction Based On Deep Learnin

With the development of the Internet and e-commerce,China has become the world’s largest online sales market,and online shopping has gradually become an indispensable way of life in the daily life of Chinese people.As an important part of enterprise management,sales forecasting allows businesses to more accurately determine production based on sales,avoid product backlogs,reasonably manage product inventory,arrange production schedules,and timely supplement products to avoid shortages and other phenomena.Compared to traditional offline shop sales,ecommerce has the advantages of easy data collection,rapid data processing and a large volume of data,which can be used to forecast merchandise sales more easily and accurately.There will be significant fluctuations in sales before and after the e-commerce organization activity day.Aiming at this feature,this paper proposes a combination model based on deep learning to study e-commerce sales forecasting,and two methods are proposed.Method 1: Using the historical sales data of an e-commerce online shop,a three-layer LSTM neural network model was built under the Keras framework.By comparing and operational results of the predictions of the CNN model,the traditional LSTM model,the combined ARMA-SVR model and the Wave Net-LSTM model,the new model has some improvements in terms of prediction accuracy and training efficiency.Method 2: Based on the research of Method 1,the Prophet model is combined with a multi-layer LSTM model in this paper,and finally the PSO algorithm is used to combine the prediction results of the two models with weights,resulting in a sales prediction network model that is more adaptable to the sales fluctuation sales characteristics of e-commerce sales before and after the event day.In addition,the sales data set of a clothing e-commerce on Taobao was used for comparison test,and the single Prophet model,multi-layer LSTM model and traditional ARIMA time series regression model were compared.The experimental results show that the combined Prophet-LSTM-PSO model has superior performance and further improvement in prediction accuracy in a situation like e-commerce where sales fluctuations are relatively high.In this paper,based on the research of e-commerce sales prediction based on deep learning,on the basis of improving the accuracy of prediction,it can effectively balance the production value and sales volume of e-commerce enterprises,have more scientific guidance for the management of inventory reserves of e-commerce enterprises,and have a better plan for the promotion of goods to promote the sales of goods.