Product Sales Combined Forecast Based On MI Algorithm

Sales forecast is of great significance for the operation of the company.Many people have proven that combined prediction is an effective way to improve prediction accuracy.However,studies have shown that the performance of a combined forecasting model derived from a combination of better-performing single-prediction models is not necessarily optimal.The MI algorithm is an algorithm for discussing information intersection and union.It is consistent with the selection of combined prediction sub-models.It has many applications in personalized recommendation,automatic text classification and key frame extraction techniques.In this paper,the MI algorithm is innovatively applied to the sales portfolio forecasting,and strives to obtain a more accurate combined forecasting model.The main contents of this paper include the following aspects:Firstly,by the tseries package,forecast package,TSA package,pdR package,numpy package,pandas package,skleam package and tensorflow package of R language and Python,ARIMA model,threshold autoregressive model,exponential smoothing model,linear regression model,Bayesian regression model,elastic network regression model,support vector regression model,gradient boosting regression model,LSTM neural network and GRU neural network are established to apply the MI algorithm.After the model is built,the predicted performance of each model is compared and analyzed.Then,each single model is added to the discussion of combined prediction,the single model selected by MI algorithm is used to establish a combined forecasting model,and the optimal single models are used to establish a combined forecasting model as a comparison.Through experiments,we find that the single model with the best performance is support vector regression model,and its RMSPE is 0.1161.RMSPE of the combined forecast model established by using the optimal single model as a sub-model is 0.1093.RMSPE of the combined forecast model established by the sub-model selected by the MI algorithm is 0.1036.The experimental results show that the combined forecast model improves the forecast accuracy,and the sub-model selected by the MI algorithm makes the prediction accuracy of the combined prediction model further improved.The MI algorithm has a good application prospect in the sales forecast research.In addition,through this experiment,the advantage of the threshold autoregressive model in short-term sales forecast and high-predictive sales forecast is also found,which proves the significance of the model in sales forecast.