Rohail-Bhana/paediatric-respiratory-ml

Dataset and code for explainable ML classification of paediatric respiratory disease (South Africa). # paediatric-respiratory-ml # Explainable Machine Learning for Paediatric Respiratory Disease Classification This repository accompanies the manuscript: > **Explainable Machine Learning Identifies Diagnostic Patterns for Paediatric Respiratory Diseases in a High-Dimensional Underrepresented South African Dataset** > Rohail Hershil Bhana, Charl Verwey, Krupa Prag, Avnish Bhowan Magan. > *[Journal name to be added on acceptance]*, 2026. DOI: *[to be added on acceptance]* It hosts the anonymised dataset and the analysis code used to benchmark and interpret machine learning models for the diagnostic classification of three paediatric respiratory conditions: asthma, bronchiectasis, and bronchopulmonary dysplasia (BPD). > **Note on availability.** This repository has been established to accompany the submission. The complete anonymised dataset and the analysis code will be uploaded here upon acceptance of the manuscript. Until then, the anonymised dataset is also provided as supplementary material with the submission. ## Overview The study presents a clinically derived, real-world paediatric pulmonology dataset captured from handwritten clinician records at Chris Hani Baragwanath Academic Hospital (Gauteng, South Africa), and benchmarks ten machine learning models for diagnostic classification under a consistent, leakage-controlled evaluation protocol. Evaluation covers discrimination (AUC), precision–recall performance (AUPRC), threshold-based metrics, probabilistic reliability (Brier score, log loss, calibration curves), and model interpretability via SHAP. ## Dataset - **Records:** 2,176 paediatric patients (earliest available consultation per patient; cross-sectional). - **Features:** 300 clinical variables — three continuous (age, weight, height) and the remainder sparse binary indicators of documented findings. - **Labels:** 95 diagnostic labels in a multi-label binary format (0 = not documented present, 1 = documented present). - **Coding convention:** unre …