Container dwell time predictive modelling: an application of ML algorithms

This study analyses factors affecting container dwell time (CDT) at the Mombasa Port using machine learning (ML) algorithms. The study employs real-time container movement data to evaluate several ML models. It finds that CDT varies significantly across different periods in the year and even in the days and weeks. For example, it peaks in the afternoons and during November/December. Although models like Artificial Neural Networks and Random Forest outperform others, the Decision Tree model was chosen for its interpretability, despite a slightly higher error rate. It identifies transportation modes as the key predictor, with truck-based movements leading to longer dwell times than rail transport. The study highlights the impact of specific locations and times of the week/year on CDT. Its originality lies in using real-time data from the Global South and its application of ML to improve operational efficiency and strategic decision-making. Unlike typical studies focused on terminal operations, this research also considers broader exogenous factors. The findings provide valuable insights for optimizing port operations and reducing CDT.