Grocery Sales Forecasting

Abstract

Forecasting grocery items can not only avoid excessive stockpiling but also meet customer demand. This can reduce the grocery store's losses and increase the grocery store's turnover. On the other hand, considering features that affect sales is also the core of feature engineering. This article makes a forecast about the sales of merchandise in a large chain store. In this paper, two cores of time series and LGBM are mainly used to complete the model establishment. A model that can predict the sales of goods is built. The data used for training is analyzed and processed. The accuracy of the model is measured using the mean squared error, which gives a final accuracy of 0.35069696616549817. At the end of the paper, it proposes an improved method for this model, and how to solve the same problem under other conditions (such as when the data is particularly small).