Improving Methods Using Machine Learning and Databases in Forensic Anthropology

Part I: GeoFOR
Estimating the time since death, or the postmortem interval (PMI), poses a significant challenge to forensic scientists when human remains are discovered because of the limited availability of reliable methods. In this presentation, we introduce geoFOR, a free web-based collaborative application and taphonomic database that utilizes geospatial mapping and machine learning to deliver improved PMI estimations.  

The geoFOR application provides a standardized, collaborative forensic taphonomy database that gives practitioners a data collection method to enter case information that automates the collection of environmental and weather data using ArcGIS and delivers a PMI estimation using statistically robust methods. The geoFOR database currently contains over 2600 entries from across the U.S. and internationally including cases from medicolegal investigations and longitudinal studies from human decomposition facilities. Thus, geoFOR provides a reference dataset that is forensically and geographically representative of the realities in which human remains are discovered. This reference dataset allows the machine learning model to continuously improve upon PMI estimations as more data is contributed by registered geoFOR collaborators.  

After case submission, the cross-validated machine learning PMI predictive model results in a R² value of 0.82. Contributors receive a predicted PMI with an 80% confidence interval. This novel method for PMI estimation can help successfully narrow the search parameters for unknown decedents which can expedite identification and more accurately inform us about the circumstances surrounding their death. 

Part II: FAST
Skeletal trauma research is critical to improve the understanding of the response of human bones to loading and the interpretation of the resulting fractures. One of the greatest limitations researchers are faced with is ensuring they can communicate their findings to practitioners as well as offer a means for their findings to be employed. The development of the Forensic Anthropology Skeletal Trauma (FAST) database was designed primarily to bridge the gap between researchers and applied professionals and students. The FAST database is a searchable database comprised of images and all test variables, with the intent of providing objective trauma interpretation for young scholars and professionals. To date, few researchers get quality hands-on training with trauma cases; even fewer have experience with known cases. Therefore, the ability for students and professionals, at all stages in their career, to be exposed to skeletal trauma with known parameters has the potential to be transformative for the field. The FAST database includes specimen, experimental, fracture characteristic, and imaging data for each record (i.e., for each skeletal element). Individual-level variables (e.g., age) and element-level variables (e.g., total length) are included in the database for each specimen. The experimental data consists of the boundary conditions (e.g., loading direction) and test output variables (e.g., force). The inclusion of these data make the FAST database unique both in the amount and quality of data within it, enabling the database to be extremely useful for both scholars and professionals in the field of forensic anthropology. 

Detailed Learning Objectives

1. Attendees will be introduced to machine learning applications in forensic anthropology and how forensic anthropological databases can support training and casework applications. 
2. Attendees will learn about the geoFOR application—a collaborative forensic taphonomic database for forensic case submission that uses a machine learning model to provide a postmortem interval estimation with associated error rates. 
3. Attendees will gain an interdisciplinary perspective of skeletal trauma through FAST by examining experimental research utilizing human specimens with known loading mechanisms.