A Multi-Layered Framework for AI-Driven Logistics Optimization: Statistical Analysis, Heuristic Modeling, and Agent-Based Decision Architecture.

This thesis develops and empirically validates a multi-layered framework for the integration of artificial intelligence and metaheuristic optimization within logistics and supply chain operations. While existing literature extensively documents the predictive capabilities of machine learning models in isolated supply chain functions, limited research addresses the structured integration of forecasting, optimization, and decision auditing within a unified operational architecture. This work bridges that gap by combining statistical analysis of AI adoption trends, a structured first-time AI implementation protocol, and a real-world case study focused on truck load planning and logistics optimization. At the operational level, the study formulates the truck load planning problem using heuristic First-Fit Decreasing algorithms and mixed-integer programming models, followed by the integration of unsupervised machine learning techniques, specifically k-means and DBSCAN clustering, to enhance shipment aggregation and routing efficiency. Comparative empirical results demonstrate improvements in load utilization, operational flexibility, and decision robustness under realistic industrial constraints. Furthermore, the thesis introduces an agent-based decision and audit layer designed to replicate and formalize dispatcher reasoning, thereby enhancing transparency and explainability in algorithmic logistics planning. The proposed architecture integrates data acquisition, optimization, clustering, external API routing, and decision auditing into a cohesive system framework. The findings confirm that the combined use of machine learning, metaheuristics, and structured governance mechanisms can significantly improve operational efficiency while maintaining managerial interpretability. This research contributes both theoretically and practically by proposing a scalable architecture that connects strategic AI adoption with executable logistics optimization in real-world environments.