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

**Authors:** Flavio Di Martino, Franca Delmastro, Cristina Dolciotti  
**arXiv:** [2305.19636](https://arxiv.org/abs/2305.19636)  
**Date:** 2023-05-31  
**Summary page:** https://www.summarizepaper.com/en/paper/2305.19636v1/

## Key points

- The paper addresses the issue of malnutrition in older populations, particularly among hospitalised or institutionalised individuals.
- Detecting malnutrition risks accurately and early is crucial for effective management and prevention.
- The authors propose an AI framework that leverages m-health data for continuous monitoring and assessment of malnutrition risks.
- Integration of AI with m-health services can significantly improve detection and management of malnutrition.
- Explainable AI (XAI) methodologies ensure interpretable and trustworthy AI decisions in healthcare settings.
- Random Forest (RF) and Gradient Boosting classifiers perform best, especially when incorporating body composition assessment data.
- Several benchmark XAI methods were investigated to extract global model explanations, with SHapley Additive ExPlanations (SHAP) showing high agreement with feature permutation method.
- Preliminary clinical validation was performed to verify the reliability and accuracy of the proposed AI framework.
- The paper presents a novel AI framework for early detection of malnutrition risks using heterogeneous m-health data.
- Combining AI technology with explainability methods offers a promising solution for improving malnutrition management in older populations.

## Summary

The paper titled "Explainable AI for Malnutrition Risk Prediction from m-Health and Clinical Data" addresses the serious issue of malnutrition in the older population, particularly among hospitalised or institutionalised individuals. Detecting malnutrition risks accurately and early is crucial for effective management and prevention. The authors propose an innovative AI framework that leverages m-health data to enable automatic, objective, and continuous monitoring and assessment of malnutrition risks. The integration of Artificial Intelligence (AI) with m-health services has the potential to significantly improve the detection and management of malnutrition. Additionally, the use of Explainable AI (XAI) methodologies ensures that AI decisions are interpretable and trustworthy for end users. This is especially important in healthcare settings where transparency and understanding are essential. The study conducted an extensive evaluation of various models, including both subject-independent and personalised predictions. The results indicate that Random Forest (RF) and Gradient Boosting classifiers perform best, particularly when incorporating body composition assessment data. Furthermore, several benchmark XAI methods were investigated to extract global model explanations. The consistency assessment revealed that each selected model prioritizes similar subsets of relevant predictors, with the highest agreement observed between SHapley Additive ExPlanations (SHAP) and feature permutation method. To validate the effectiveness of their approach, a preliminary clinical validation was performed to ensure that the learned feature-output trends align with current evidence-based assessments. This step verifies the reliability and accuracy of the proposed AI framework. In conclusion, this paper presents a novel AI framework for early detection of malnutrition risks using heterogeneous m-health data. By combining AI technology with explainability methods, this framework offers a promising solution for improving malnutrition management in older populations. The findings highlight the importance of incorporating body composition assessment data and demonstrate the consistency between different XAI methods in providing meaningful explanations for model decisions.