Research On Commodity Sales Forecast Based On Tree Model

The 21st century is the era of big data.Machine learning and data mining are widely used in many fields such as retail,medical,and transportation.With the gradual maturity of big data storage platforms such as Hadoop and Spark,the storage of business data and customer data of various retail enterprises is no longer a problem.How to analyze and mine these data and turn them into valuable information that can bring benefits to the company is a challenge for all enterprises.Under the "new retail" environment,the competition among various retail enterprises has turned into data competition and technology competition.Big data and related technologies have effectively promoted the development of the retail industry.As an important part of the retail industry,merchandise sales can accurately predict it with the help of machine learning,which can help companies fight well in today's changing business situation and enhance their core competitiveness.Until today,many small-scale stores still take human observation and decisionmaking for the sales of goods.For companies that apply data mining prediction methods to commodity sales,the commonly used prediction method is the time series method.But this method has certain disadvantages.Because it only depends on the time feature and does not extract the data,so to a certain extent,a lot of information is wasted in the data,and there are still many valuable information that has not been mined.However,algorithms like XGBoost and LightGBM,which are very popular in data mining competitions,can solve this problem well,and fully mine the potentially valuable information in the data.Only accurate sales forecasts can bring better decisions to decisionmakers and establish a good relationship with customers.Inaccurate forecasts will cause out-of-sale products to be sold out and unsalable goods to accumulate,resulting in the loss of customers,which is not conducive to the development of store companies.In order to build a more accurate prediction model,this article will use the integrated model based on the decision tree to study the real historical sales data of the store,conduct a detailed visual analysis of the original data,explore the valuable information in the data,and build a random forest.,XGBoost,LightGBM and other prediction models,analyze the characteristics of each model,and compare the prediction effects of each single model one by one.In order to make the prediction accuracy more accurate,the model fusion is improved on the basis of stacking,and this method is effective after experimental verification.