Back to Projects

AI For Search and Rescue

Machine learning models achieving 80%+ accuracy in predicting missing person patterns to aid search and rescue operations.

86%recall, wander-status model
75%precision, wander-status model
2conference presentations
15person ML team led
PythonScikit-learnK-Nearest NeighborsK-meansSVMPandasExplainable AIISRID

About the Project

Led a 15-person machine learning team on IntelliSAR, a Cal Poly research collaboration with experienced search and rescue leaders under Dr. Franz Kurfess. We trained K-means, K-Nearest Neighbors, and SVM models on the International Search and Rescue Incident Database (ISRID), Robert Koester's historical missing-person dataset, to surface behavioral patterns that narrow a search area. The K-Nearest Neighbors wander-status model reached 86% recall and 75% precision on a heavily imbalanced dataset where only 14% of incidents ended unsuccessfully. The work identified predictive relationships between elevation, subject situation, and rescue outcome, and between age, gender, and time missing. Presented twice at the National Missing and Unidentified Persons Conference in Las Vegas, most recently in April 2025.

Impact

Research presented at the National Missing and Unidentified Persons Conference (Las Vegas) twice

Project Presentation

Project Documentation