Demand Forecasting Model Development using Machine Learning: Case of Mongolian Retail Company

Abstract

Because of the rapid development and expansion of mobile and e-commerce, the demand of retail and logistics industry has been greatly increased. In addition, customers gets the opportunity to purchase a lot of stuffs through the integrated channels both online and offline. However, this trend makes retail companies have difficulties to prepare more products and control the inventory. Especially, it gets more important to predict future demand. But, the life cycle of product gets shorten, thus it is impossible to predict demand based on the long-term historical data. In order to overcome the limitations of the traditional demand forecasting method, the cluster based demand forecasting methods have been proposed. Still, the previous research could not solve the limitations because they utilized the input variables from the categories or specifications of product. In this research, we have proposed the different approach to utilize the meta-data which can describe the sales patterns. Based on these pattern, we developed the cluster of products which are categorized into different groups. After integrating the sales data, we have developed demand forecasting models using deel learning technology, LSTM.