Container Demand Forecasting Using Machine Learning Methods: A Real Case Study from Turkey

Abstract

The container demands in ports significantly fluctuate over time and accurate container demand forecasting is essential for logistics companies because they can make their future business plans accordingly. In maritime transportation, container slot agreements are generally made two times in a year. A slot is one Twenty-Foot Equivalent Unit (TEU) space in a container ship and early booking of a slot is less costly for a company. Therefore, the accurate prediction of future container demands is crucial for companies to reduce their costs and increase their profits. In this study, we developed various forecasting models using machine learning methods to accurately predict the future container demands for the largest maritime transportation and logistics company of Turkey. The main aim is to provide accurate container demand forecasts for the company so that it can optimize the container slot bookings. To forecast the container demand, we used the company`s internal demand data as well as various external data such as gross domestic product (GDP), inflation rate, and exchange rate. We built four forecasting models based on Linear Regression, Boosted Decision Tree Regression, Decision Forest Regression and Artificial Neural Network Regression algorithms. The performances of these methods were evaluated according to Coefficient of Determination, Mean Absolute Error, Root Mean Square Error, Relative Absolute Error and Relative Squared Error. The case study showed that Boosted Decision Tree Regression and Decision Forest regression methods yield the best forecasting accuracy.