MACHINE LEARNING-BASED BODY MASS INDEX (BMI) CATEGORY CLASSIFICATION USING ANTHROPOMETRIC AND LIFESTYLE FEATURES FOR ATHLETE HEALTH ASSESSMENT

Abstract

Body Mass Index (BMI) is a widely used indicator for assessing general health and body composition; however, its applicability in athletic populations is limited due to its inability to distinguish between fat mass, lean mass, and sport-specific physiological adaptations. This study proposes an interpretable machine learning-based framework for BMI category classification using anthropometric and lifestyle-related features. The data involved height, weight, training age, training duration and how often meals are taken per day. Since there are few observations (about 100 samples) available, a controlled data augmentation method was used, which employs Gaussian perturbation to increase the dataset size to 500 samples with real physiological relationships. The BMI values were re-calculated and divided into four groups namely; Underweight, Normal, Overweight and Obese. An 80:20 traintest split was used to train a Decision Tree classifier which was run in the scikit-learn library of Python. An overall classification accuracy of the model was about 94-95%. The confusion matrix analysis revealed that the majority of samples were properly classified with slight misclassification between similar categories like Overweight and Obese. The analysis of feature importance showed that height and weight were the most significant predictors, and lifestyle-related variables played a less significant role. The findings prove the usefulness of interpretable machine learning methods in BMI classification and indicate their possible use in athlete health assessment. Future research is required on larger datasets and more complicated modeling approaches to enhance generalization.