Supervised machine learning algorithms for the classification of obesity levels using anthropometric indices derived from bioelectrical impedance analysis

The accurate classification of obesity is essential for public health and clinical decision-making. Traditional anthropometric measures such as body mass index (BMI) have limitations in differentiating between fat and lean mass. This study aimed to evaluate and compare the performance of various supervised machine learning algorithms in classifying obesity levels using anthropometric indices derived from bioelectrical impedance analysis (BIA). A cross-sectional study was conducted on a sample of 5372 adults (age 34.6 ± 10.0 years) (2727 females and 2645 males). Anthropometric data included BMI, fat mass index (FMI), fat-free mass index (FFMI), skeletal muscle index (SMI), muscle mass index (MM), and others were collected using a validated multifrequency octopolar BIA device (InBody 270). Six supervised machine learning models, random forest, gradient koosting, k-nearest neighbors, logistic regression, support vector machine, and decision tree, were trained and evaluated using accuracy, precision, recall, F1-score, area under the receiver operating characteristic curve (AUC-ROC), and SHapley Additive exPlanations value explanations. Random forest outperformed all other models, achieving the highest accuracy (84.2%), F1-score (83.7%), and AUC-ROC (0.947). SHapley Additive exPlanations analysis revealed that FMI, FFMI, and BMI were the most influential features, while sex had minimal predictive impact. Machine learning models, particularly tree-based algorithms like random forest, show great potential in classifying obesity levels from anthropometric data with high accuracy and interpretability. These models can enhance the effectiveness of obesity screening in clinical and community settings.