A review of the literature on the applications of machine learning in forensic anthropology

This is the authors' abstract. We don't add key points for this paper.

Applications of machine learning (ML) models in forensic anthropology have increased in the last half decade. It is, therefore, important to understand the context in which machine learning models are being used in this discipline. The aim of this paper is to provide the current state of machine learning applications in forensic anthropology through a systematic review process of the literature. This paper provides a descriptive summary of existing literature, rather than a deep critical analysis of the methodological robustness. The literature search was performed using Scopus and Web of Science from 1987 to 2024. Eligible studies were investigated if they had a forensic anthropological focus with an application of machine learning. A total of 167 papers were analyzed after the exclusion criteria were applied. The results of this paper demonstrate that there is a wide range of machine learning model applications in forensic anthropology, utilizing diverse bones, applied to all aspects of the biological profile and some aspects of trauma analysis. Through this review, it is also encouraged to use ML models on underutilized skeletal elements to optimize the pattern recognition capabilities of machine learning for validation of forensic anthropological assessments. Validating ML models on less commonly analyzed skeletal elements will increase the skeletal elements that can be utilized to assist in identifying and repatriating individuals. The review also demonstrates that there is an increased need to provide a comprehensive description of the machine learning applications to increase transparency, interpretability, and further validation in forensic anthropological assessments. • A total of 167 papers were identified to have used machine learning models for forensic anthropological assessments. • Machine learning models were used mostly for biological sex assessments (n= 79 studies). • Neural networks (n = 42 studies), and combinations of different ML models for comparative studies (n = 91 studies) were the most common approaches. • The skull was the most common skeletal area used for ML applications (n = 52 studies), followed by the pelvis (n = 23 studies).