Sales Forecasting During the COVID-19 Pandemic for Stock Management

Abstract

AbstractStock management is very important for the companies to supply necessary demand for the products they sell, to pricing the products and to the aspect of storage cost. In stock management, the products to be sold to the customer are procured by ordering from the vendors. The orders given to the vendors are determined by estimating the sales quantities of the products. When estimating sales, if we order in large quantities, the storage and expiration dates of the products may exceed, or if we order less than demand, the customer cannot find the product in the store. With the Covid-19 pandemic entering our lives, there have been some changes in our habits. One of these changes is the change in shopping habits of people due to the isolation period. By managing this change in terms of stock management on the store side, it ensures that people can reach the products they demand in these difficult times and that companies do not create extra costs by making more stock than necessary. We made a sales forecast on 5-lt sunflower oil which is a basic food product using the data of a grocery chain with machine learning methods and developed models to use these forecasts in stock management. Our data is multivariate and contains both quantitative and qualitative features. In our study, we used the supervised learning method and the XGBoost, LGBMRegressor and Ridge models used in many machine learning projects. As a result of our studies, an improvement of approximately 25% has emerged with the features we added specifically for the pandemic.