Machine learning-based comparative analysis of nasal morphology: Forensic implications for sexual dimorphism and ethnic variation

Abstract

The human nose is a prominent facial feature whose morphoscopic traits offer valuable insights into an individual's ancestry, sex, and personal identification. The study aims to evaluate the nasal morphological features to explore sexual dimorphism and ethnic differences among the two North Indian population groups using three machine learning models: Random Forest Classifier (RFC), CatBoost (CB), and Support Vector Machine (SVM). Snapshots of the noses of 500 participants (18–35 years) from frontal, basal, and lateral profiles belonging to two genetically distinct endogamous populations (i.e., Rajputs and Brahmins) from North India, were collected. A total of eighteen morphoscopic parameters such as nose size, nasal bridge profile, etc. were considered and evaluated through standard procedures. Data were pre-processed and underwent a single train-test split into 80:20 ratio along with a fivefold cross-validation method. Customised machine learning models predicted sex and ethnicity, with SHAP (SHapley Additive exPlanations) used for interpretability, while intra-observer reliability was assessed to ensure data consistency. Support Vector Machine performed slightly best for sex estimation (69%), followed by Random Forest (68%) and CatBoost (62%). For ethnicity, Random Forest performed slightly better, achieving 75% accuracy, followed by CatBoost (72%) and Support Vector Machine (70%). SHAP analysis revealed nasal root height and nose size as consistent sex predictors, whereas ethnicity classification relied on a more variable traits across models. Moreover, fivefold cross-validation results provided an assessment of model performance stability. Nasal morphoscopic parameters show greater ethnicity than sex, underscoring their forensic utility in sex and ethnicity estimation.