Related Experiment Video
Updated: Aug 12, 2026
13:19
Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
Deep learning approaches show promise for predicting childhood malnutrition: A comparative study with traditional
Deepak Bastola1,2, Yang Li2
1Texas College of Management and IT, Kathmandu, Nepal.
Plos One
|July 22, 2026
Summary
Machine learning models can identify childhood malnutrition in Nepal. TabNet performed best, highlighting maternal education and wealth as key predictors for targeted interventions.
Area of Science:
- Public Health
- Computational Biology
- Data Science
Background:
- Childhood malnutrition is a significant public health issue in Nepal and other low-resource regions.
- Traditional methods for identifying malnutrition are often resource-intensive and inaccessible in remote areas.
Purpose of the Study:
- To apply machine learning (ML) and deep learning (DL) for identifying child malnutrition in Nepal.
- To compare the performance of 16 different ML and DL algorithms.
Main Methods:
- Utilized data from the Nepal Multiple Indicator Cluster Survey (MICS) 2019.
- Developed a composite malnutrition indicator integrating stunting, wasting, and underweight status.
- Systematically evaluated 16 algorithms, including deep learning, gradient boosting, and traditional ML, focusing on F1-score and recall.
Main Results:
- TabNet, an attention-based deep learning model, demonstrated the highest performance among all evaluated algorithms.
- Key predictors for malnutrition included maternal education, household wealth index, and child age.
- Geographic factors, vaccination status, and meal frequency also emerged as significant predictors.
Conclusions:
- The study presents a scalable, survey-based framework for screening children at risk of malnutrition.
- This approach can guide targeted nutritional interventions and support progress towards Sustainable Development Goals.
- The methodology offers a transferable template for similar low-resource settings globally.