A hybrid AI framework for predictive delay optimization and smart freight dispatching in maritime and intermodal logistics

The accelerating complexity of freight logistics, driven by dynamic operational variables and high customer expectations, has necessitated the integration of intelligent decision-support systems. This study presents a dual-layered artificial intelligence framework that combines supervised learning and reinforcement learning to enhance delay prediction and dispatch optimisation in logistics. Using the Smart Logistics Supply Chain Analysis dataset from Kaggle, a gradient boosting classifier (XGBoost) was trained to predict shipment delays based on features such as inventory level, shipment status, traffic conditions, and environmental variables. The model achieved perfect classification performance, with an F1-score, precision, and recall of 1.00, and an AUC of 1.00, highlighting its ability to generalise across complex operational patterns. Feature importance analysis revealed that inventory pressure, shipment phase, and traffic congestion were the most influential delay factors. Complementing this predictive layer, a reinforcement learning (RL) agent was developed to simulate intelligent dispatch decisions. The RL agent showed increasing cumulative rewards across training episodes and selected deliveries uniformly, demonstrating fairness and adaptability. Data visualisations, including pairplots, ROC curves, and dispatch distribution maps, provided interpretive depth and operational relevance. This hybrid AI framework bridges the gap between foresight and control in logistics operations. It offers a deployable blueprint for freight platforms seeking to reduce delays, improve resource allocation, and respond dynamically to real-world uncertainties.