Total retail sales forecasting method based on CEEMDAN, GPR combined model and CBR correction

Abstract

The total retail sales of social consumer goods is an important index reflecting the consumption level. Predicting its development trend is helpful to grasp the economic development situation. Because the factors affecting the total retail sales of social consumer goods are very complex, it is very difficult to make full use of the value information contained in a single forecasting model. Therefore, this paper proposes a new forecasting method of total retail sales of social consumer goods corrected by combined model and case-based reasoning (CBR). First, a new dataset is constructed by using adaptive noise complete set empirical mode decomposition (CEEMDAN) to eliminate high frequency noise. The differential integration autoregressive moving average (ARIMA), long and short term memory (LSTM), limit gradient enhancement (XGBoost) and support vector regression (SVR) models were established for the new dataset, and then the prediction results of each prediction model were integrated with Gaussian process regression (GPR) to obtain the initial prediction value and error sequence. In addition, in order to solve the problem that the implicit knowledge in the error sequence is difficult to be regularized and quantified by mathematical model, this paper proposes a new error correction method, namely CBR, to improve the prediction accuracy. Experimental results show that compared with single model, the method proposed in this paper has better prediction effect and can effectively improve the prediction accuracy of total retail sales of social consumer goods.