Early Detection of Common Infant Illnesses (0–3 Years) Using Explainable Machine Learning: A Symptom-Based Decision Support System for Caregivers

Background: Early identification of common infant illnesses is essential for reducing diagnostic delays and improving child healthcare outcomes, particularly in resource-limited settings where timely access to pediatric specialists may be challenging. Objective: This study proposes an explainable machine learning framework for the early prediction of common infant illnesses using symptom-based and demographic information collected from Bangladesh. Methods: A clinically validated dataset comprising 2,552 infant health records representing eight disease categories was developed from hospital records and caregiver reports. The dataset was divided using a stratified 80:20 train–test split, while five-fold cross-validation was employed during hyperparameter optimization to prevent overfitting and ensure robust model selection. Four supervised machine learning algorithms, namely Logistic Regression, Decision Tree, Random Forest, and Extreme Gradient Boosting (XGBoost), were trained and evaluated. Randomized Search CV with five-fold cross-validation was employed to optimize the best-performing model. Model robustness was further assessed using bootstrap confidence intervals, calibration analysis, and an ablation study, while explainability was achieved through feature importance analysis and SHAP. Results: Among the evaluated models, the optimized Random Forest classifier achieved the highest predictive performance with an accuracy of 99.22%, a Macro F1-score of 99.04%, and a best cross-validation Macro F1-score of 0.9907. Bootstrap analysis demonstrated narrow 95% confidence intervals, and the average Expected Calibration Error was 0.0097, indicating excellent model stability and probability calibration. The ablation study further confirmed that incorporating demographic information alongside symptom variables improved predictive performance. Conclusions: These findings demonstrate that the proposed framework provides accurate, robust, and interpretable predictions and has the potential to support caregivers and primary healthcare professionals in the early recognition of common infant illnesses while promoting trustworthy artificial intelligence for pediatric clinical decision support.

Authors

Institutions

Publication Details

Journal
Journal of Emerging Global Health
Published
2026-11-21
DOI
https://doi.org/10.36481/cynq5729
Primary Topic
Artificial Intelligence in Healthcare
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Early Detection of Common Infant Illnesses (0–3 Years) Using Explainable Machine Learning: A Symptom-Based Decision Support System for Caregivers

Naznin Sultana, Radoanul Arifen, S.M Meriyan Islam
Journal of Emerging Global Health
Artificial Intelligence in Healthcare
article

Early Detection of Common Infant Illnesses (0–3 Years) Using Explainable Machine Learning: A Symptom-Based Decision Support System for Caregivers

Naznin Sultana, Radoanul Arifen, S.M Meriyan Islam
article en

Abstract

Background: Early identification of common infant illnesses is essential for reducing diagnostic delays and improving child healthcare outcomes, particularly in resource-limited settings where timely access to pediatric specialists may be challenging. Objective: This study proposes an explainable machine learning framework for the early prediction of common infant illnesses using symptom-based and demographic information collected from Bangladesh. Methods: A clinically validated dataset comprising 2,552 infant health records representing eight disease categories was developed from hospital records and caregiver reports. The dataset was divided using a stratified 80:20 train–test split, while five-fold cross-validation was employed during hyperparameter optimization to prevent overfitting and ensure robust model selection. Four supervised machine learning algorithms, namely Logistic Regression, Decision Tree, Random Forest, and Extreme Gradient Boosting (XGBoost), were trained and evaluated. Randomized Search CV with five-fold cross-validation was employed to optimize the best-performing model. Model robustness was further assessed using bootstrap confidence intervals, calibration analysis, and an ablation study, while explainability was achieved through feature importance analysis and SHAP. Results: Among the evaluated models, the optimized Random Forest classifier achieved the highest predictive performance with an accuracy of 99.22%, a Macro F1-score of 99.04%, and a best cross-validation Macro F1-score of 0.9907. Bootstrap analysis demonstrated narrow 95% confidence intervals, and the average Expected Calibration Error was 0.0097, indicating excellent model stability and probability calibration. The ablation study further confirmed that incorporating demographic information alongside symptom variables improved predictive performance. Conclusions: These findings demonstrate that the proposed framework provides accurate, robust, and interpretable predictions and has the potential to support caregivers and primary healthcare professionals in the early recognition of common infant illnesses while promoting trustworthy artificial intelligence for pediatric clinical decision support.

Journal of Emerging Global HealthVol. 13(1)
Daffodil International University (BD)
Artificial Intelligence in Healthcare
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon