Machine Learning Models for Prediction of Sex Based on Lumbar Vertebral Morphometry.

Area of Science:

• Forensic Anthropology
• Bioinformatics
• Machine Learning

Background:

• Skeletal remains identification is challenging, particularly in mass disasters or decomposed remains.
• Forensic experts and anthropologists face difficulties in sex determination from skeletal evidence.

Purpose of the Study:

• To assess the reliability of machine learning (ML) techniques for sex prediction using L1-L5 lumbar vertebrae morphometry.
• To determine if metric analysis of the lumbar spine can provide accurate sex identification in forensic contexts.

Main Methods:

• Developed and tuned predictive models using Random Forest (RF) and XGBoost (XGB) classification techniques.
• Employed cross-validation and grid search for hyper-parameter optimization.
• Selected optimal models based on ROC_AUC (area under the curve) performance metric.

Main Results:

• L1-L5 lumbar vertebrae exhibit sexual dimorphism, serving as effective predictors for sex determination.
• RF model identified six significant predictors, while XGB model highlighted three key predictors for sex prediction.
• Both algorithms demonstrated reliable performance in predicting sex based on lumbar spine measurements.

Conclusions:

• RF and XGB techniques reliably predict sex using L1-L5 measurements, even with a small dataset.
• This ML approach is valuable for forensic identification when only skeletal remains are available.
• The ML setup can be adapted into an accessible web service for forensic anthropologists to predict sex from lumbar spine data.