Research On Sales Prediction Of Chain Drug Stores Based On Ensemble Learning

In the inventory optimization management of the pharmaceutical industry,the sales forecast of chain drugstores has potential application.However,chain pharmacy sales forecasting is complex in terms of data quality and time series features.There are still significant limitations in forecasting using traditional ensemble learning.This research aims to combine the characteristics of chain pharmacy retailing,explore and improve the shortcomings of the original ensemble learning in this application scenario.This thesis conducts in-depth research and analysis on the sales forecasting literature,focusing on its ideas,implementation methods and applications,and provides theoretical support for the realization of an accurate and stable sales forecasting model for chain pharmacies.The main contents of this study include the following three aspects:(1)Collect and analyze two sales data sets.These include a drug pharmacy sales data set covering a domestic listed chain pharmacy company,and a Rossmann chain pharmacy sales data set on the Kaggle platform.Perform data exploration and data preprocessing.First,analyze the characteristics of basic data distribution and sales trends.Then,perform preprocessing operations such as filling and encoding on the basic data.(2)Model optimization based on Light GBM.By observing the data structure and practical application characteristics of chain pharmacy sales forecasting,and introducing neural network to improve the original Light GBM,a self-attention mechanism and Light GBM fusion sales forecasting model(TS-LGBM)was proposed.TS-LGBM extracts sales timing parameters by introducing a self-attention mechanism,which can better fit time series predictions.At the same time,the synthetic sales feature of the sliding window is used to further improve the time series expression ability of the model.(3)Model prediction effect analysis.Four models,TS-LGBM,TS-XGB,Light GBM,and XGBoost,were successively built,and experiments were conducted on two chain pharmacy sales data sets.The comparative experimental results show that the prediction accuracy of the TS-LGBM model is better than the other three models.To sum up,this study is oriented to practical engineering applications.Through accurate chain pharmacy sales forecasting,data can be effectively used and data value displayed,supply chain management efficiency can be improved and costs can be reduced.Research has strong application.