Handicrafts Sales Trend Prediction Based On Wavelet Analysis And ARMA-SVR

With the increasingly prosperous in handicrafts trading industry and the development of e-commerce,a large number of handicrafts online trading platforms have been arisen.These online trading platforms have accumulated a large number of users’ browsing data,the page views or clicks on handicrafts can reflect the users’ needs more accurately,which can provide an objective and realistic reference for the improvement of the marketing strategy,and has a positive effect on handicrafts sales.We can mining and analysis the browsing data by using the advanced technology of data mining and get the handicrafts sales trend.Then a reliable support for the sales decision can be provided.Based on the analysis of the current handicrafts sales trend prediction,this paper uses the technology of web crawler and web information extraction to acquisite the browsing data of the typical handicrafts online trading platforms,then a time series of page views has been constructed.Then the problem of predicting the sales trend can be transformed into a nonstationary time series predicting problem.In view of the fact that the prediction accuracy of single model is not high,the technology of wavelet analysis is introduced.Then the original time series formed from page views of handicrafts on trading platforms are decomposed and reconstructed by Mallat algorithm,several high frequency signals and a low frequency signal are gotten.The high frequency signals are predicted with the model of ARMA,and the low frequency signal is predicted with the model of SVR.The final prediction result of the original series is the superimposition of these respective prediction results.The experimental results show that the prediction precision of the model based on combination of wavelet analysis and ARMA-SVR is 10.28%higher than the model based on wavelet analysis and ARMA,and 49.97%higher than the model based on wavelet analysis and SVR.The prediction precision of the model based on ARMA or SVR without using wavelet analysis is very poor and can’t reflect the characteristics of the original time series well.The results show that the method proposed in the paper can get better prediction accuracy for the series of page views and can be used in handicrafts sales short-term trend prediction.