Facial-Based BMI Estimation for Digital Health Monitoring Using MediaPipe and Meta-Ensemble Learning

The estimation of Body Mass Index (BMI) plays a critical role in assessing an individual's health and risk factors for various diseases. Traditional methods for BMI calculation rely on height and weight measurements, which may not always be accessible, such as in telemedicine and remote health monitoring. In this work, we propose a novel approach to predicting BMI using facial images and machine learning techniques. The methodology incorporates facial landmark extraction, feature selection, and regression-based BMI estimation. The Illinois dataset, consisting of facial images and corresponding BMI values, is used to train and evaluate the model. Mediapipe's Face Mesh model is employed to extract geometric facial features, followed by a meta-ensemble model for BMI prediction. Experimental results show that the proposed approach achieves significant results with an RMSE of 1.9976, MAE of 0.9600, Pearson correlation of 0.9254, and R² score of 0.8554.