The Enhancement Algorithm And Comparative Study Of Sales Forecasting Model Based On Data Mining Technology

With the rise and wide application of artificial intelligence algorithm,more and more commercial enterprises begin to attach importance to the combination of their own business with big data,artificial intelligence and other technologies,and look forward to using emerging technologies to improve the company's operation process,improve the company's performance and promote the development of enterprises.The application and research and development of big data,artificial intelligence and other technologies have gradually become an important part of today's enterprise operation and management.In the operation and management of an enterprise,sales volume forecast is a crucial link.Reasonable sales volume forecast can accurately predict the future performance of an enterprise based on historical data and provide an important basis for its operation and decision-making.According to sales forecast problem,we generally refer to the relevant research literature at home and abroad,and combining experience design their own study plan: we studied the integration model of tree XGBoost,LightGBM and random forest algorithm,analyses their features,and dedicated to a single model and mixed model algorithm is enhanced,and then compared the advantages and disadvantages of these algorithms are theoretically,and through the sales forecast experiment from two aspects of forecasting accuracy and model training time to explore and improve the performance of the algorithm model.We used the data set of Kaggle's Sales forecast competition Rossmann Store Sales to conduct modeling experiments.Firstly,the original data is analyzed,and the corresponding preprocessing and feature construction are carried out.Then,the characteristics of XGBoost,LightGBM and random forest are compared and analyzed in detail,and an algorithm enhancement strategy composed of result correction and multi-model combination is proposed.The last three kinds of algorithms and enhancement algorithm using Python3.6 modeling experiment.The results show that,in the single algorithm model,the prediction accuracy of XGBoost model is the highest,while the training time of LightGBM and random forest model is faster,and the prediction accuracy of random forest model is the lowest.The result correction in the enhancement strategy can greatly improve the prediction accuracy of the model.On the basis of result correction,the model combination enhancement strategy can further improve the prediction accuracy of the model.In the combination model,the multi-algorithm combination model including XGBoost,LightGBM and random forest has higher prediction accuracy than the combination model with a single algorithm.The training time of the combined model is shorter than that of the XGBoost combined model.This shows that the proposed multi-model combination strategy improves the performance of the algorithm from the two aspects of prediction accuracy and model training time.