Decision trees for estimating osteological sex from the skull using an expanded suite of morphological traits.

Abstract

Osteological sex estimation utilizing morphological traits of the skull primarily focuses on a set of five traits. Developing new models with an expanded suite of traits has the potential to increase sex classification accuracies and improve the classification of partial and fragmentary remains. Thus, this study seeks to improve classification accuracies for osteological sex estimation from the skull by assessing an expanded suite of traits and generating multiple decision tree classification models. Twenty-one traits were evaluated for a sample of 403 adult males and females. Krippendorff's alpha was used to assess intraobserver error, and aligned rank transformation was used to examine the effects of sex, age, population, and secular change on the traits. Prior to generating the decision trees, the data were randomly split into 80% model training samples and 20% holdout validation testing samples to test the predictive accuracy of each tree. The trees were generated for traits from the skull, cranium, and mandible. Separate trees were also generated for African Americans and European Americans, as well as for the pooled population sample. Overall, the recommended decision trees for the skull and cranium achieved higher accuracies (91.0%-100%) than models generated for the mandible (75.8%-85.0%). Additionally, the recommended pooled population (81.3%-97.3%) decision trees achieved similar accuracies compared with the African American (75.8%-94.0%) and European American (85.0%-100%) trees. Further, the trees generated in this study achieved improved classification accuracies for the skull compared with the current five-trait method by incorporating an expanded suite of traits.