Enhancing Search and Rescue: A Data-Driven Approach to Abduction Case Analysis and Predictive Modeling

Abductions are a persistent and dangerous issue around the world, forcing search and rescue (SAR) teams to make rapid decisions to effectively allocate limited resources. The goal of this project is to improve SAR operations by identifying the most probable recovery locations and critical case characteristics. This project was completed in collaboration with dbS Productions LLC, who provides SAR teams with research, publications, software, training, and maintains the International Search and Rescue Incident Database, a key resource in lost person behavior research. The project builds on this foundation by developing statistical models to assist SAR teams in identifying high-priority search areas and predicting key case outcomes. The methodology consists of several key steps, from data cleaning, synthetic data generation, ring model generation, and application of Bayesian techniques.Our results indicate that Age and Homicide Status are statistically significant predictors in determining an abducted person’s distance from the Initial Planning Point. Additionally, our synthetic data maintained similar distributions to the original data, addressed class imbalance, and proved to be robust when tested in an analytical setting. We created and tested a number of models that built upon existing SAR models and achieved better results across the board when testing with the MapScore metric. Our Bayesian model demonstrated the best performance, achieving a MapScore of 0.701.