Research On Store Sales Forecast Based On BP Neural Network And KNN

With the rapid development of the Internet age,more and more people choose to shop online.Compared with traditional shopping,people can enjoy the convenience and fun of online shopping without leaving their homes,which significantly saves time and improves efficiency.At the same time,it also promotes the continuous development and increase of online shops.The e-commerce industry began to occupy the market,and the online shop has become a new business model.It is difficult to solve financing problems for e-commerce companies.With high financing costs and reasonable lending,it has become a kind of commercial consensus.Therefore,scientific forecasting of its store sales status makes it reasonable for operators of online stores to have a realistic guiding significance.This thesis supports the sales of certain online stores in a network mall from 2016 to 2017 to ensure the authenticity and accuracy of the experimental results.According to the actual application,the influencing factors of store sales are analyzed.Based on the existing data results,the BP neural network and K-nearest neighbor algorithm are used to analyze and be compared with the store sales forecasts,and the accuracy of model application is studied.Finally,the two models are combined.Abnormal data make more reasonable predictions.The results of analysis error show that BP neural network and K nearest neighbor algorithm have their own advantages and disadvantages.Studying the two models on this issue has a certain contrast effect and can be aimed at the actual situation and make a reasonable prediction.In the thesis of store sales forecasting,store sales are forecasted based on indicators including the daily orders,sales,customers,evaluations,and advertising costs of several stores.Based on the analysis of the BP neural network,the data training will be used for the first ten days to predict the sales for the next three days.It is determined that the neural network input layer,hidden layer,and output layer are 40,20,and 3,respectively.The forecast time is about 5 minutes.In KNN prediction model,according to the existing experience,by continuously changing the size of K value in order to reduce the error to determine the K value of 3,the prediction time is about 10 minutes.In the absence of abnormal data,the prediction accuracy of BP neural network is better than that of KNN.When there is abnormal data,KNN performs better than BP neural network.Through the analysis and comparison of the two model prediction results,KNN algorithm is used to process the anomalous data in advance,and then it is predicted by the BP neural network.Finally,the results of the two predictions are weighted and averaged to obtain the final result.The final prediction result indicates that the algorithm of the two models combined(abbreviated as BP-KNN in the text)has a better effect on abnormal data.