A case study on automatic reordering of perishable products based on time series forecasting

Abstract

Sales forecasting is a key element in the large-scale retail trade. A good forecasting system is usually connected to a (semi) automatic reordering platform and in general to the inventory information system of the company. The following work aims at providing an overview of the results of several forecasting models. The case study is built around the request of a company in the large-scale retail trade which requests a system for automatic reordering of perishable food and short shelf-life food. The nature of the perishable products implies that standard techniques, such as the stochastic-service approach and the guaranteed service, are not an option because they required a longer shelf life to works in a reasonable way to reduce cost and limit waste. As a consequence, time-series analysis and forecasting were required. Both classical methods, like ARIMA and ETS, and most recent techniques based on a combination of statistical methods and neural networks were tested and applied. Common evaluation metrics assume symmetric errors and do not consider the economic evaluation, a new metric has been proposed to overcome these limitations.