# Sex classification using exocranial surfaces in a multi-population sample

## TL;DR

This study evaluates a new method for determining sex from skull surfaces, showing high accuracy in some populations but lower reliability in others.

## Contribution

The study introduces a reliable sex estimation method using exocranial surfaces, validated across multiple populations.

## Key findings

- The method achieved 96% accuracy for Czech and 92% for Slovak populations.
- Accuracy dropped to 82% for Egyptian and 80% for Danish populations.
- The classifier's reliability decreases when applied to geographically distant or diverse populations.

## Abstract

The accurate individual identification of skeletal remains is indispensable in forensic contexts. The skull serves as an important source of information about the sex of human skeletal remains, and many different approaches have been published. High method success and reliability are prerequisites for the legal utilisation of results. However, the population specificity of variable sexual dimorphism typically reduce effectiveness. This study presents a verification of an innovative classification model using the exocranial surface across a multi-population sample. This sex estimation method proved to be highly reliable and accurate for Central European populations, achieving high accuracy rates for Czech (96%) and Slovak (92%) samples. The French sample had an accuracy of 90%, demonstrating the method’s effectiveness in Southern European contexts. Prediction using the combined data from these three populations achieved a cross-validation accuracy of 91.74%. When this classifier model was applied to Egyptian crania, the accuracy dropped to 82%, and when applied to crania from a Danish dataset to 80%. The reasons for the failure of the classifier are the smaller degree of sexual dimorphism among Danes, and the more distinct morphological differences in males and females among Egyptians. These lower accuracy rates indicate that the classifier’s reliability diminishes when applied to more diverse and geographically distant populations. The classifier does not work well when applied to a population other than that for which it was developed. The method is robust, and requires further refinement to achieve similar reliability across a broader range of populations.