Research on commodity sales forecasting based on combination model

In the age of Internet, accurate sales forecasting model is of great significance to help enterprises cope with the changes of market demand and make a reasonable inventory plan. Due to the influence of many non-linear factors, the precision of traditional sales forecasting model is often unsatisfactory. In order to improve the precision of sales forecasting, this paper uses PCA to reduce the dimension of the input LSTM network data features and retain the useful information in the data, taking full advantage of the advantages of XGBoost model and LSTM neural network model, after removing the noise data, the PCA_LSTM-XGBoost combined sales forecasting model is constructed based on the entity historical data, and the comparison experiments with the single model and two typical time series forecasting models are designed and implemented. The experimental results show that under the same experimental conditions of the enterprise, the PCA_LSTM-XGBoost combination prediction model can reduce the MAPE of at least 8% in the commodity sales volume prediction compared with other models, which indicates that it has higher accuracy and better applicability.