An Automated Multi-Output Framework for Shipment Rerouting and Delay Prediction in Supply Chain Networks

Abstract

Worldwide freight ecosystems process enormous shipment volumes every year, yet a considerable share is affected by transit setbacks and inefficient path planning, resulting in elevated expenses and reduced service reliability. Existing management practices largely depend on human supervision or rigid rulebased systems, which struggle to adapt to rapidly changing conditions and fail to provide timely responses. To address this gap, this work introduces a data-centric predictive framework aimed at early identification of delivery disruptions and enhancement of operational efficiency. The approach utilizes a rich dataset containing shipment schedules, route histories, asset utilization information, and feasibility indicators. A structured preparation pipeline involving cleansing, scaling, and transformation is applied to ensure high-quality inputs for modeling. For comparison, benchmark techniques such as K-Nearest Neighbor with Classification and Regression Tree (KNN-3CA1RT) and Huber-3CA1RT are analyzed. The proposed solution employs a Decision Tree-driven hybrid 3CA1RT (DT-3CA1RT) architecture that integrates three classification stages alongside a regression component. These modules sequentially detect rerouted consignments, assess resource suitability, and validate operational feasibility, while the regression unit estimates expected delay durations. Experimental evaluation confirms that the proposed model delivers superior predictive capability and supports more effective decision-making. By automating disruption detection and feasibility validation, the framework enables proactive planning and minimizes operational inefficiencies. Additionally, the system is implemented as a Flask-based application, allowing users to input shipment details, generate instant predictions, and optimize routing and resource distribution, thereby improving overall system performance and dependability.