A Sequence-Based Neural Network Model for Obesity Prediction among Adults Aged 18-64 Based on Physical Fitness Variables

Abstract Obesity is a global health concern. We developed an explainable sequence deep learning model using data from 108,304 adults (2010–2023) to predict obesity. Among five stratified folds, Fold 4 was optimal (TN=1,462; FN=184; FP=249; TP=1,554), yielding 0.873 accuracy (precision=0.866, recall=0.855, F1=0.876, ROC-AUC=0.95) and stabilizing at 20 epochs. Its MAE=0.122, MSE=0.041, R²=0.833, and average prediction error=0.171 for the first 50 samples. SHAP analysis revealed 20-m VO₂ max (importance=0.339), gender (0.2481), and grip strength (0.135) as top predictors; LIME (intercept=0.511, predicted=0.668, R²=0.995) indicated that low 20-m Shuttle Run VO₂ max (<28.71 ml/kg/min) and grip strength (<38.17%) strongly increased obesity risk, especially among females. Overall, the model achieved 0.87–0.88 accuracy (p<0.001 vs. random), underscoring the vital role of cardiorespiratory fitness in managing obesity.