Naranjo-Zolotov, Mijail JuanovichAcedo Sánchez, AlbertReyes Zerené, Nicolás Esteban2025-11-172025-10-31http://hdl.handle.net/10362/190836Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics, specialization in Business AnalyticsNon‑pickups (nPU) in business‑to‑business vehicle‑transfer services remain a critical source of operational cost and reputational risk. This thesis augments earlier spatial analyses with advanced feature engineering and ensemble learning to improve the early detection of nPUs. After auditing column types and harmonising five years of booking records (2019–2024), we engineered lead‑time windows and multi‑scale seasonality indices. The enriched dataset fed a hybrid modelling pipeline that combines Multi‑scale Geographically Weighted Regression (MGWR) for interpretability with XGBoost for predictive accuracy. Preliminary analyses indicate an uplift in predictive performance, while MGWR maps still expose supplier‑capacity hotspots in Southern Europe. The study demonstrates that spatial awareness and tailored temporal features jointly advance proactive mitigation strategies for logistics providers.engNon-Pickups (nPU)MGWRB2B Vehicle TransfersPredictive ModellingSpatial AnalysisSupplier Performance OptimizationOperational Risk Mitigationmaster thesis204073260