Study of different data science methods for demand prediction and replenishment forecasting at retail network

Abstract

The demand prediction becoming an essential tool to remain or even lead in the competition among the retail businesses. A well-done demand prediction model could help retailer to track the level of inventory, orders and sales in the most effective way in which the best results could be achieved. However, there are many different methods and opinions of how to create a demand prediction model. In this paper, we will analyse the most commonly used methods of Linear regression, Logistic Regression, Probabilistic Neural Network, Bayesian Additive Regression Trees, Random Forest and Fuzzy Logic with their specifications and limitations found in studies of authors. After review performed all methods will be compared according to characteristics selected. Moreover, in order to get more practical results the accuracy of Logistic Regression and Random Forest methods will be compared based on data of milk sales collected from retail network. For constructing of decision support system for retail network, we need to go beyond demand prediction one-step to replenishment forecasting. It was concluded that there is no best method to forecast replenishment and results can differ based on the data and conditions analysing. In every situation authors seeking to select the method with the highest accuracy and the lowest number of errors possible. Limitations of research: limited number of goods and stores included in the modelling.