An Interpretable and Generalizable Machine Learning Model for Predicting Asthma Outcomes: Integrating AutoML and Explainable AI Techniques

Asthma remains a prevalent chronic condition, impacting millions globally and presenting significant clinical and economic challenges. This study develops a predictive model for asthma outcomes, leveraging automated machine learning (AutoML) and explainable AI (XAI) to balance high predictive accuracy with interpretability. Using a comprehensive dataset of demographic, clinical, and respiratory function data, we employed AutoGluon to automate model selection, optimization, and ensembling, resulting in a model with 98.99% accuracy and a 0.9996 ROC-AUC score. SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-Agnostic Explanations) were applied to provide both global and local interpretability, ensuring that clinicians can trust and understand model predictions. Additionally, counterfactual analysis enabled hypothetical scenario exploration, supporting personalized asthma management by allowing clinicians to assess potential interventions for individual patient risk profiles. To facilitate clinical adoption, a Streamlit v1.41.0 application was developed for real-time access to predictions and interpretability. This study addresses key gaps in asthma prediction, notably in model transparency and generalizability, while providing a practical tool for enhancing personalized care. Future research could expand the validation across diverse patient populations to reinforce the model’s robustness in broader clinical environments.