Enhancing sex determination in forensic anthropology: A comparative analysis of cranial measurements using artificial neural network

Abstract

The increasing reliance on machine learning (ML) techniques in forensic anthropology underscores the imperative to enhance the accuracy and objectivity of sex estimation from skeletal remains. Traditional methods often suffer from subjectivity and variability, prompting a shift towards morphometric approaches for improved precision. In this context, this study aims to identify the most accurate cranial measurements for sex estimation and ascertain the minimum number of variables needed to match the performance of a 12-measure model using a neural network. The objective of this study was to determine the most accurate individual cranial measurements for sex estimation and to identify the minimum number of variables required to achieve comparable accuracy to the 12-measure model using a neural network. Data from 241 skulls from the collection of the Center for Studies in Anatomy and Forensic Anthropology at the University of Tiradentes were used in this study. Twelve measurements were performed. The data were divided into a test group (20 %) and a training group (80 %). Machine learning algorithms were developed using the Python language in the Google Colaboratory environment. The combination of three measurements (biporion, head circumference and opisthocranion-glabella) outperformed the reference linear models and the 12-measure model in the neural network. Biporion isolated in a neural network outperformed 12 measures in logistic regression. Our study adds to the literature a high accuracy with only three measures, reducing the workload for the examiner, in a reproducible and reliable way, using neural networks.