• Surveyed 1,000 rural households in Mahendragarh (Haryana) and Jhunjhunu (Rajasthan) to assess health impacts of biomass fuel use. • Identified high prevalence of respiratory symptoms, especially among women and children exposed to smoke from wood, dung cakes, and crop residues. • Developed a comprehensive analytical system integrating PM monitoring, weather data, and health surveys to model exposure-risk relationships. • Applied machine learning (Random Forest, XGBoost, LSTM) and causal inference methods (Granger causality, Causal Impact) to predict pollution and health outcomes. • Designed real-time API and dashboard tools for public health alerts and policy support in rural air quality management. This study investigates the spatiotemporal dynamics of ambient air pollution and its health impacts across two semi-urban districts in India- Jhunjhunu and Mahendragarh, using a multidisciplinary approach combining statistical analysis, machine learning, and causal inference. A one-year high-resolution monitoring dataset of PM₁, PM₂.₅, PM₄, and PM₁₀ was integrated with structured household health surveys covering over 1,000 households. High-resolution monitoring of PM₁, PM 2.5 , PM₄, and PM₁₀, along with survey-based health data, was analyzed to explore pollutant behavior, exposure-response relationships, and symptom prevalence. Linear regression models effectively predicted PM 2.5 trends in Jhunjhunu, while advanced models such as Random Forest, XGBoost, and Long Short-Term Memory (LSTM) captured complex variability in Mahendragarh. Models were trained using a 70:30 train–test split with k-fold cross-validation and evaluated using RMSE, MAE, and R² metrics. LSTM and XGBoost achieved the best performance (R² up to 0.87; RMSE reduced by approximately 30% compared to linear regression). SHAP analysis highlighted PM₁ and PM₄ as critical predictors, underscoring the need to expand national air quality standards beyond PM 2.5 and PM₁₀. Explainable machine learning using SHAP identified PM₁ and PM₄ as influential predictors of health-related outcomes, underscoring the need to expand national air quality standards beyond PM2.5 and PM₁₀. Granger-causal links, residual diagnostics, and health symptom anomalies revealed significant associations between particulate pollution and respiratory, cardiovascular, and visual symptoms, particularly in Mahendragarh. Policy insights emphasize cleaner fuel adoption, improved ventilation, and awareness campaigns to mitigate risk among vulnerable, low-income households. By integrating machine learning with epidemiological modeling, this study provides robust, location-specific evidence to support targeted environmental health interventions in under-monitored regions. A key innovation of this study lies in the joint monitoring and modeling of PM₁ and PM₄ alongside conventional PM₂.₅ and PM₁₀ using explainable ML and causal inference. This framework captures nonlinear exposure–response patterns and improves predictive accuracy while providing mechanistic insight into particle-size-specific health risks. The results offer actionable evidence for clean fuel transition, household ventilation improvements, and community-level air quality management in semi-urban and rural settings. Integration of high-resolution particulate monitoring, machine learning, and causal inference reveals strong links between PM₁-PM₁₀ exposure and cardiopulmonary and ocular symptoms in semi-urban India, highlighting PM₁ and PM₄ as key predictors for targeted interventions.
KUMAR et al. (Sun,) studied this question.