Estimation of body mass index using a machine learning method based on hand and foot anthropometric data

In forensic contexts, particularly in cases involving dismembered or severely decomposed remains, biological profiling may rely on limited anatomical information. This study aimed to evaluate the feasibility of body mass index (BMI) classification using machine-learning (ML) methods based exclusively on hand and foot anthropometric measurements.

A total of 287 healthy individuals aged 18-25 years were prospectively included. Anthropometric measurements of hand and foot length and circumference were obtained using standardized methods. Because the underweight and obese categories contained relatively few participants, ML analyses were restricted to 240 individuals in the normal-weight and overweight categories (120 participants per group). Eleven ML models were compared, and their performance was evaluated using accuracy, precision, recall, F1-score, and ROC-AUC. The dataset was split into 80% training and 20% testing sets, and five-fold cross-validation was applied.

The Random Forest model achieved the highest performance (accuracy = 0.87; F1 = 0.87; AUC = 0.92), followed by XGBoost and Gradient Boosting. All anthropometric parameters showed significant positive correlations with height and body weight (p < 0.001). Circumferential measurements were more strongly associated with body weight, whereas length measurements were more strongly correlated with height. Significant sex-related differences were observed across all parameters (p < 0.001).

These findings provide preliminary proof-of-concept evidence that hand and foot anthropometric measurements may support ML-based discrimination between normal-weight and overweight individuals. However, the restricted BMI categories and absence of external validation limit the generalizability of the findings, and further validation in larger, more heterogeneous, and postmortem populations is required before forensic applicability can be established.