Combining morphological traits and measurements of the skull for osteological sex estimation using random forest modeling

Forensic anthropologists commonly estimate osteological sex using separate morphological and metric analyses, without integrating both data types into a single classification model. Combining data types into one model has the potential to increase sex classification accuracies for the skull. Therefore, the present study seeks to improve classification accuracies for the skull by combining morphological and metric variables using random forest (RF) modeling. The main objectives are (1) generate multiple RF models that incorporate various combinations of morphological and metric variables for estimating osteological sex from an unknown individual, (2) compare the performance of morphological, metric, and combined data RF models, and (3) compare the results of the RF models to current methods for osteological sex estimation of the skull. The sample included 212 European Americans (males = 106, females = 106) and 191 African Americans (males = 114, females = 77). The models were trained on 80% of the sample and tested using a 20% holdout sample. Multiple models were generated using morphological, metric, and combined variables. Across all model types, the skull and cranium models achieved higher accuracies compared to the mandible models. The morphological and combined models attained higher accuracies compared to the metric models. Additionally, the morphological and combined RF models attained comparable classification accuracies to current standard osteological sex estimation methods, as well as compared to previous studies that integrated skull measurements and traits. Future research should continue exploring RF modeling for osteological sex estimation, including models combining metric and morphological variables from multiple skeletal regions.