Explainable Artificial Intelligence for Predicting Calorie Imbalance

This study demonstrates the potential of machine learning in predicting calorie imbalance by integrating physiological, behavioral, and dietary features into an interpretable analytical framework.Using Logistic Regression and Random Forest models, the results showed that both approaches achieved high predictive accuracy, with Logistic Regression slightly outperforming Random Forest in overall classification metrics.The findings indicate that even relatively simple and transparent models can capture the complex interplay between energy intake and expenditure when supported by well-prepared and meaningful data.The SHAP analysis provided valuable insights into the relative importance of individual variables, highlighting that calorie intake, calories burned, workout type, and session duration were the most influential determinants of energy balance.These factors emphasize the pivotal role of daily activity patterns and dietary habits in shaping caloric equilibrium.Meanwhile, demographic attributes such as age and gender contributed less significantly, suggesting that lifestyle behavior exerts a stronger and more direct effect on calorie imbalance.For future research, it is recommended to expand the dataset with more diverse demographic and geographic samples to improve generalisability across populations.Incorporating longitudinal data would also allow for temporal tracking of energy balance dynamics rather than single-point estimation.Furthermore, integrating additional behavioural and psychological variables-such as stress level, motivation, or eating context-may enrich model interpretability and predictive depth.Finally, future studies could explore the deployment of these interpretable models in mobile health platforms, enabling realtime monitoring and personalised feedback for users aiming to maintain energy balance in daily life.