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

Aquatic Ecology · Springer Nature · 2026-11-12 · DOI: 10.1186/s41072-026-00265-z

Аннотация

The increasing complexity of freight logistics, driven by fluctuating operational variables and high customer expectations, has led to the adoption of intelligent decision-support systems. This study introduces a two-layered AI framework that integrates supervised learning and reinforcement learning to improve delay prediction and dispatch optimization in logistics. Using the Kaggle Smart Logistics Supply Chain Analysis dataset, a gradient-boosted classifier (XGBoost) was trained on pre-dispatch operational and environmental attributes, strictly excluding post-hoc status variables to prevent target leakage. Under 5-fold stratified cross-validation, the model achieved a realistic ROC-AUC of 0.7831 ± 0.0220, an F1-score of 0.7309 ± 0.0304, and a Precision of 0.8876 ± 0.0409. Feature importance analysis identified traffic congestion, demand forecast, inventory pressure, and location factors as key contributors to delays. Additionally, a reinforcement learning (RL) agent was created to simulate smart dispatch decisions, achieving increasing rewards over training episodes and fairness in delivery selection. Visual tools such as pair plots, ROC curves, and dispatch maps added interpretive value and operational insight. This hybrid AI approach effectively links prediction with control in logistics, providing a ready-to-deploy blueprint for freight platforms aiming to reduce delays, optimize resource use, and adapt to real-world uncertainties.

Читать у издателя →

Похожие статьи