Sex estimation from long bones: a machine learning approach.

Area of Science:

• Forensic Anthropology
• Bioinformatics
• Biometrics

Background:

• Sex estimation from skeletal remains is vital in forensic anthropology.
• Long bones offer an alternative for sex estimation when primary methods fail due to degraded or absent remains.
• Machine learning presents a novel approach to enhance accuracy in skeletal sex estimation.

Purpose of the Study:

• To develop and compare machine learning classification models for sex estimation.
• To evaluate the efficacy of using combined measurements from multiple long bones.
• To identify the most accurate machine learning model for sex estimation from skeletal data.

Main Methods:

• Collected eighteen measurements from four long bones (radius, humerus, femur, tibia) from 2141 individuals.
• Employed five machine learning classifiers: Linear Discriminant Analysis (LDA), Penalized Logistic Regression (PLR), Random Forest (RF), Support Vector Machine (SVM), and Artificial Neural Network (ANN).
• Utilized cross-validation techniques to assess model performance on a validation sample.

Main Results:

• Classification models using all combined long bones achieved high accuracy, ranging from 90% to 92% on the validation sample.
• Models using isolated bones showed accuracy between 83.3% and 90.3%.
• The Random Forest (RF) model demonstrated the highest accuracy, proving most effective for this study's objectives.

Conclusions:

• Combined measurements of long bones provide highly accurate models for sex estimation.
• Machine learning, particularly Random Forest, offers a robust and accurate method for sex estimation in forensic anthropology.
• The developed models are applicable across diverse populations, enhancing their forensic utility.