Cross-Domain Data Integration and Hybrid Learning for Resilient Forecasting in Intelligent Logistics Ecosystems

Abstract

Smart logistics systems increasingly depend on accurate demand forecasting to optimize global supply chain operations. Recent industry reports indicate that logistics bottlenecks impact over 30% of global shipments, while inaccurate forecasting contributes to nearly a 25% rise in operational costs. Traditional manual estimation methods are often insufficient, as they fail to incorporate dynamic real-world factors such as fluctuating traffic conditions, IoT-generated data, and environmental variations. This study presents a robust data fusion framework that integrates IoT, traffic, and meteorological datasets to enhance demand forecasting and logistics delay prediction. The methodology begins with a comprehensive preprocessing phase, where heterogeneous data sources are cleaned, normalized, and temporally synchronized to ensure consistency and reliability. Baseline models such as K-Nearest Neighbors (KNN), Classification and Regression Tree (CART), and Categorical Boosting (CB) are employed to provide initial predictive insights; however, these models often struggle to capture complex interdependencies within multi-source data. To address this limitation, a Tree-based Adaptive Optimization Tree (TAO Tree) model is proposed, designed to improve learning from heterogeneous features. This design enables accurate regression for demand prediction and effective classification for logistics delay detection. The system is deployed through a Flask-based web application, enabling realtime data processing, prediction, and visualization. Experimental results demonstrate that the proposed framework significantly improves forecasting accuracy and operational efficiency, offering a scalable and data-driven solution for proactive supply chain management.