Demand Forecasting Using Machine Learning to Manage Product Inventory for Multi-channel Retailing Store

Nowadays, there are many small and medium retailing stores that have multi-distribution channels on both online and offline platform. Each channel is also divided into several sub-channels so that it may affect the store's inventory management system. Some of the product types or quantities may not be sufficient to meet the needs of customers or some products may exceed the customer demand. These cause the cost of transportation, storage, labor, and most importantly, the store may lose its income and profit. To help those retailing stores to manage their inventory effectively, a method for forecasting customer demand by using machine learning to manage product inventory for multi-channel retailing stores is presented in this paper. We use the product sales data from one of the retail stores in Thailand as a reference. They are the daily sales report of each item in the year 2017 to 2021 retrieved from each branch of the store. The dataset comprises 178,548 rows, that are used to create the demand forecasting models for 7-day and 30-day periods. This paper mainly focuses on creating a demand forecasting model by using CatBoost algorithm. The performance is compared to the models that are generated by XGBoost algorithm and Linear Regression algorithm. According to the evaluation of the efficiency, the SMAPE of the 7-day demand forecasting is 24.13% and the SMAPE of the 30-day demand forecasting model is 24.47%.