public health informatics

Ambient air pollution is a major driver of respiratory, cardiovascular, and metabolic disease in India, yet frameworks that jointly address forecasting, prediction, uncertainty, and interpretability remain scarce in low- and middle-income countries (LMICs). We propose the Hybrid Explainable Air–Health Learning (HEAL) framework, a four-stage pipeline combining (i) Prophet–LSTM pollutant forecasting, (ii) ensemble machine learning with the Synthetic Minority Over-sampling Technique (SMOTE), (iii) Bayesian calibration, and (iv) Explainable Boosting Machines (EBM). Data from the Longitudinal Ageing Study in India (LASI; n = 43,267) were linked with Central Pollution Control Board air quality records to predict six outcomes: COPD, asthma, cardiovascular disease, stroke, type 2 diabetes, and lung cancer (prevalence 0.9–6.8%). The hybrid forecaster achieved the lowest error (RMSE 13.2 vs. 14.8 for LSTM and 16.5 for Prophet). HEAL outperformed baseline models for all six outcomes (AUC 0.58–0.64), and SMOTE improved minority-class recall by 11–18 percentage points, with the largest gains for the rarest diseases. Bayesian calibration yielded narrow 95% credible intervals, and bootstrap sensitivity analysis (±20% pollutant perturbation) confirmed stable associations. Age was the dominant predictor, while O3 and NO2 were most influential for respiratory outcomes and particulate matter for cardiometabolic outcomes. HEAL offers an accurate, interpretable, and uncertainty-aware approach to multi-disease environmental health surveillance in LMIC settings. The specific objectives of this study are to forecast ambient air pollutant concentrations using a hybrid time-series approach combining Prophet and Long Short-Term Memory (LSTM) models; to improve prediction of multiple pollution-related health outcomes using ensemble machine learning with the Synthetic Minority Over-sampling Technique (SMOTE); to quantify predictive uncertainty through Bayesian calibration by estimating credible intervals for individual-level risk predictions; and to enhance interpretability using Explainable Boosting Machines (EBM) to identify disease-specific pollutant effects and interaction patterns. The Longitudinal Ageing Study in India (LASI) is a household survey of the elderly, which is nationally representative and organized by the International Institute of Population Sciences (IIPS), Mumbai [14]. The basis year (2017-2018) itself comprised 43,267 people aged 45 years and above and represented all 28 states and 8 union territories in India. The LASI dataset gives self-reported physician-diagnosed chronic conditions, such as asthma, chronic obstructive pulmonary disease (COPD), chronic bronchitis, lung cancer, cardiovascular disease, diabetes, stroke, and tuberculosis.