Explainable AI for Malnutrition Risk Prediction from m-Health and Clinical Data

AI can improve malnutrition detection in vulnerable populations: The framework successfully identified malnutrition risk in older adults using mobile health data and clinical information, with Random Forest and Gradient Boosting models performing best when body composition data was included—potentially enabling earlier intervention in hospitals and care facilities.

Explainability methods show consistent agreement on what matters: Multiple explanation techniques (SHAP and feature permutation) identified similar key predictors for malnutrition risk, with "the highest agreement shown between SHAP and feature permutation method," suggesting these explanations are reliable enough for clinicians to trust and act upon.

Clinical validation is essential for responsible AI deployment: The researchers verified that the AI model's learned patterns aligned with established medical evidence, demonstrating that explainability alone isn't sufficient—AI predictions must be validated against real-world clinical knowledge before use in patient care.

Personalized predictions may outperform one-size-fits-all models: The study evaluated both "subject-independent and personalised predictions," suggesting that tailoring AI models to individual patients could improve accuracy and potentially lead to more appropriate, individualized malnutrition interventions.

Transparency builds trust in healthcare AI: By making AI decisions interpretable through explainability methods, the framework addresses a key barrier to adoption—clinicians and patients are more likely to trust and use AI tools when they can understand why a malnutrition risk was flagged rather than treating it as a "black box."