Explainable Deep Learning-Based BMI Estimation from Eating Habits and Physical Activity Patterns

As an important public health crisis, obesity is associated with unhealthy lifestyle factors, like eating habits and physical activity patterns. It is related to an increased risk of various chronic diseases. Therefore, accurately estimating body mass index (BMI) is essential for identifying at-risk groups and supporting early interventions. This study proposes an Explainable Deep Learning-Based BMI Estimation (XDL-BMI) framework based on eating habits and physical activity characteristics. The presented model aims to estimate BMI by identifying the features that contribute to the model’s BMI predictions. To achieve this, the proposed model initially processes the input data to ensure reliable model development. Following that, the Minimum Redundancy Maximum Relevance method is employed to select the relevant features, and a Deep and Cross Network is designed to estimate BMI by learning the relationships among the selected factors. Finally, the SHAP method is integrated to enhance the model’s interpretability by identifying factors that most influence BMI estimation. To assess the efficiency of the proposed model, an experimental analysis is carried out using the UCI Estimation of Obesity Levels Based on Eating Habits and Physical Condition dataset. The experimental results of the proposed model achieved a normalized MSE of 0.0022, RMSE of 0.0464, MAE of 0.0360, MAPE of 0.0684, and R2 of 0.9136 on the testing set. After inverse transformation to the original BMI scale, the model achieved an RMSE and MAE of 2.3562 kg/m2 and 1.8312 kg/m2, demonstrating its effectiveness in BMI estimation and providing interpretable insights into the contribution of eating and physical-activity-related factors.