Product Demand Forecasting Based on LGBM Deep Learning Models

Abstract

As the first line of defense of the enterprise supply chain, product demand forecast plays an important role in enterprises in different industries, so it is of research value to forecast it accurately.This paper preprocesses the product demand data set of a large domestic manufacturing enterprise, and through the exploration of data characteristics, it is concluded that seven different variables have a certain influence on product demand.Using feature engineering, a new column is constructed, and it is coded by single heat and map mapping.By using lag features and window statistics, 49 data features are constructed and screened.Four machine learning models, such as XGBoost, LightGBM, decision tree regression, and random forest, are constructed respectively, and the parameters are compared vertically with the grid parameters.Three indexes, mean square error (MSE), root mean square error (RMSE) and determinable coefficient (R2) are selected to evaluate the model.Accurate estimation can reduce the inventory cost of enterprises and make a pricing scheme with higher information content.