Lo, Ling I Martin; (2024) Applying Machine Learning Methods to Age and Sex Estimation in Forensic Anthropology. Doctoral thesis (Ph.D), UCL (University College London).

Text Ling I Martin Lo - PhD Thesis.pdf - Accepted Version Access restricted to UCL open access staff until 1 September 2025. Download (7MB)

Abstract

The application of machine learning and artificial intelligence within the forensic science discipline has increased in recent years. The rise of machine learning focused research has led to concerns and discourse over algorithmic opacity and transparency issues in governmental reports and literature. The application of new technologies in forensic anthropology is still relatively new, with their potential and effects still unexplored. This thesis seeks to explore uses of machine learning methods and medical images at the intersection of forensic anthropology, virtual anthropology, and 3D forensic science. A multi-output pre-processing pipeline is presented. This pipeline facilitates the extraction of digitized resources from skeletal databases such as those created in hospitals, streamlining tasks such as dataset manipulation, thresholding, and noise removal. Using data generated from the pipeline, this research explores machine learning methods for age and sex estimation by employing a novel voxelfeature approach with three-dimensional image files. The findings of the Decision Tree, Random Forest, and Convolutional Neural Network studies demonstrate the feasibility of using alternative data and new approaches in forensic anthropology. Furthermore, the machine learning studies highlight the importance of establishing fit-for-purpose datasets for better algorithmic performances. While the developed models are not yet suitable for direct applications, the research offers a step towards automating identification using new technologies. Moreover, the work aligns with current global concerns, providing insights to age estimation methods for the living. Through the empirical work carried out, this thesis raises questions of transparency in forensic science with regard to machine learning and artificial intelligence, specifically addressing how concepts of algorithmic interpretability and explainability intersect with the forensic science process. The contributions of this thesis provide timely insights for advancing research and practical applications using machine learning and medical images to forensic anthropology and its adjacent fields.

Download activity - last month

Export as CSVJSONXML

Download activity - last 12 months

Export as CSVXMLJSON

Downloads by country - last 12 months

Export as CSVXMLJSON

Archive Staff Only

View Item