Abstract

PM<inf>0.1</inf> from incense burning represents a critical indoor air quality indicator that significantly impacts health, well-being, and productivity, particularly for vulnerable occupants such as elderly individuals and patients with respiratory conditions. However, PM<inf>0.1</inf> concentrations exhibit dynamic fluctuations influenced by complex environmental factors, including building characteristics, incense stick variations, and seasonal effects, rendering traditional prediction approaches inadequate for real-time monitoring and interpretation. This research addresses these challenges by developing an interpretable machine learning (ML) framework that integrates ensemble learning algorithms with SHAP interpretability analysis, utilizing readily available ambient parameters—temperature, relative humidity, wind speed, and PM<inf>2.5</inf>—as proxies for indirect measurement of environmental influences. Experimental validation was conducted using a controlled chamber setup measuring PM<inf>0.1</inf> concentrations across varying incense stick quantities representative of religious practices in Southeast Asia, encompassing diverse building characteristic configurations. Comparative analysis through forward-chaining cross-validation with Bayesian optimization for hyperparameter tuning demonstrates that the optimized LightGBM model significantly outperforms standard ML algorithms, achieving superior predictive performance (R² = 0.8322) with robust error metrics (MAE = 1.7411 × 10³, RMSE = 5.6904 × 10³). SHAP interaction analyses reveal complex nonlinear relationships and threshold-dependent behaviors between indoor environmental variables, providing transparent explanations that enable building occupants to understand key factors influencing PM<inf>0.1</inf> concentrations and implement interventions to prevent undesirable conditions affecting health and well-being. This ensures that the interpretable framework can be effectively deployed in intelligent indoor environmental monitoring systems for proactive air quality management.