SPATIAL ANALYSIS AND OPTIMIZATION MODELS FOR BETTER UNDERSTANDING OF ILLICIT DRUG SUPPLY CHAINS WITH SPARSE AND INCOMPLETE DATA

Abstract

Illicit drug trafficking networks are difficult to observe directly due to their clandestine nature, resulting in sparse, incomplete, and often inconsistent data. These limitations hinder law enforcement efforts, including identifying trafficking routes, detecting emerging hot spots, and understanding the underlying structure of illicit supply chains. This dissertation develops methodological approaches to improve the analysis and inference of cocaine trafficking networks under such data constraints. Using cocaine-related data from Colombia and the United States, the research applies geographically weighted regression analysis, introduces an improved spatial data analysis method, and develops a data-driven optimization framework to generate actionable insights for law enforcement and policymakers. The first study investigates region-specific drivers of cocaine trafficking in Colombia using a local geographically weighted regression framework, addressing spatial heterogeneity, sparse observations, and separation issues. The results reveal substantial geographic variation in the factors associated with cocaine trafficking destinations, demonstrating that the determinants of trafficking activity differ considerably across regions. The second study examines the use of Local Indicators of Spatial Association (LISA) for identifying trafficking hot spots. Using crack cocaine seizure data from the United States, the study shows that conventional LISA methods may overlook meaningful spatial patterns in zero-heavy datasets. An enhanced LISA framework is proposed to better capture local clustering under extreme sparsity, resulting in the identification of hot spot patterns that differ meaningfully from those obtained using traditional approaches. The third study develops a data-driven optimization model to infer hidden cocaine flow networks in the United States from incomplete and noisy observations, including seizure records, undercover purchases, and overdose data. The proposed framework constructs plausible trafficking networks consistent with observed data and allows the exploration of alternative network structures through adjustable design parameters. Validation against anecdotal trafficking intelligence demonstrates the capability of the approach for identifying potential supply chain structures. Collectively, these studies contribute new methodologies for analyzing illicit supply chains in settings characterized by limited information, spatial heterogeneity, and partially observed network structures. The findings demonstrate how statistical and optimization-based approaches can enhance understanding of cocaine trafficking systems and support more effective interdiction and enforcement strategies.