Finding a Needle in the Haystack: Predicting the Location of Lost People Using Agent-Based Modeling and Behavioral Inertia

Abstract

Around 100,000 persons go missing annually in the US. Many factors go into predicting the location of a lost person, like geography, climate, age, health status, gender, disabilities, walking speed, and more. Technology and machine learning can advance the success and speed of search and rescue (SAR) missions. Agent-based modeling is a popular method for predicting the location of a lost person. A recently published paper by Hashimoto et al. demonstrated the ability to tune an agent-based model's parameters so that its emergent behavior statistically matches the lost hiker data in International Search and Rescue Database.Hashimoto et al.'s work assumed that a lost person randomly selects a reorienting behavior at every time step. We build upon the work performed by Hashimoto et al. by adding the concept of inertia as a parameter. We hypothesize that a lost person will likely continue their reorienting behavior for some time before changing. We used International Search and Rescue Incident Database SAR incidents with geolocation and Sava et al.'s scoring methodology to compare our inertia-enabled agent-based model with Hashimoto et al.'s model and validate our results. We find our model outperforms Hashimoto et al. within our testing set.