Title: Container freight rate level forecasting with machine learning methods
Authors: Yaling Feng
Addresses: School of Logistics Engineering, Shanghai Martime University, Pudong New Area, Shanghai, 201306, China
Abstract: In an ocean container transportation market, the fluctuation of spot rate is volatile, making it difficult for market players to optimise freight related operations. This paper applies an multivariant approach to first identify the significant factors affecting rate development and then utilises machine learning methods to handle the complex multivariant nonlinearities, attempting to achieve an balance between forecasting accuracy and economic factor interpretation. In the Far East-Europe trade case study, it is found that four key factors, demand, the fitting value of ARIMA model, utilisation rate and containership fleet growth, affect the rate significantly; and among the tested machine learning methods, random forest outperforms the other. Comparing with the best univarite model (ARIMA), the obtained multivariant model performs equally well, indicating that despite of the complex nonlinearities between the variables in container freight, multivariant modelling in combination with machine learning methods has the potential for reliable prediction.
Keywords: container freight rate; multivariant forecasting; univariate forecasting; machine learning.
DOI: 10.1504/IJBPSCM.2022.125689
International Journal of Business Performance and Supply Chain Modelling, 2022 Vol.13 No.3, pp.245 - 263
Received: 01 Sep 2021
Accepted: 22 Dec 2021
Published online: 26 Sep 2022 *