Combination Model In The Total Number Of The Mobile Contacts Forecasting

With the development of mobile Internet,mobile phone has become the main source of people’s access to information,and has become an advertising terminal for advertisers.In recent years,the trend of user research and advertising cost cutting is growing,and the demand for advertisers changing from violence to digital delivery is increasing too.In order to make up for the timeliness,economic problem and the low efficiency of the traditional channel investigation,the total number of the catalyst is predictability,convenience and economy,so the number of the catalyst is predicted by the Reach or the exposure quantity.In order to make better use of the forecast value of the total number of media contacts to solve the commercial problem,this paper finds a more suitable method to predict the total number of the media contacts by establishing a single model and a combination model.First,we try 5 time series models and 1 regression models,such as ARIMA(2,1,0)(0,1,0)[7],and then,from easy to difficult to combine 6 single models.Finally,comparing the root mean square error of single prediction and combination prediction,the prediction method of ARIMA-SVM is selected,that is,to fit the nonlinear signals that can not be extracted by ARIMA through SVM.Part of the interest,so as to achieve the improvement of prediction accuracy.In the process of research,it is found that the ARIMA-SVM combination forecasting method is different from the weight based combination method,which can not only guarantee low training error but also guarantee low test error,but the combination method with weight as the core is often higher than a few single model with higher prediction precision.Besides,the weight normalization method is more suitable than the method of weight reunification.In terms of selecting models,two models are selected based on the significance of the model coefficients,the trend of prediction and the accuracy of prediction.In the face of business problems,if the coefficients of the model are accepted under alpha =0.1,then the model ARIMA(2,0,0)(0,1,1)[7]+SVM(6)can be used to predict.If the coefficients of the model are only significant under the alpha =0.05,then the prediction value is needed to be processed when the ARIMA(2,1,0)(0,1,0)[7]+SVM(6)model is applied.Although this paper predicts the total number of mobile phone contacts from sixty-ninth to seventy-fifth days,the total number of the media contacts in other channels is needed on the one hand,and on the other hand,the average,maximum or trend is needed according to the customer’s demand.At the end of this paper,we summarize the problems encountered in the application of the method to the total number of the media contacts and the previous literature,and write clearly the steps of this article,and the other work and exploration that have not been mentioned in the article.