Novel Data Analytics Meets Conventional Container Shipping: Predicting Delays by Comparing Various Machine Learning Algorithms
Adrian Viellechner, Stefan Spinler
Abstract
Supply chain disruptions are expected to significantly increase over the next decades. In particular, delay of container vessels is likely to escalate due to rising congestion from continued growth of container shipping and higher frequency of extreme weather events. Predicting these delays could result in significant cost savings from optimizing operations. Both academic research and container shipping industry, however, lack analytical solutions to predict delay. To increase transparency on delay, we develop a prediction model based on 315 explanatory variables, 10 regression models, and 7 classification models. Using machine learning algorithms, we obtain best results for neural network and support vector machine with a prediction accuracy of 77 percent compared to only 59 percent of a naive baseline model. Various shipping players including sender, carrier, terminal operator, and receiver benefit from the easy-to-use prediction model to optimize operations such as buffers in schedules and the selection of ports and routes.