AI-enhanced prediction of respiratory irritation from biomass combustion byproducts: An integrative machine learning framework for sustainable bioenergy systems

Abstract

Biomass burning has continued to be a part of renewable energy infrastructure, but due to incomplete oxidation, they produce fine particulate matter and volatile organic compounds that cause pulmonary oxidative stress in the exposed populations. The study is a dual-paradigm machine learning model that combines the frameworks of both the Random Forest and the Long Short-Term Memory to forecast respiratory health risks of bioaerosol exposure in biomassdependent rural areas. Environmental monitoring data (PM2.5 concentrations, nitrogen oxides, and volatile organic compounds) in time synchrony with the clinical data (inflammatory biomarkers) were integrated to predict against clinical biomarkers to create real-time risk stratification predictive models. Nitrogen oxides were the most important feature according to the importance analysis, which confirmed mechanistic connections between inefficiency in combustion and respiratory inflammation. Random Forest model was the best in the interpretability with low prediction bias whereas LSTM networks performed high-fidelity in their ability to model temporal exposure dynamics. This integrative framework closes the environmental engineering, medical diagnostics, and optimization of renewable energy, provides evidence-based opportunities in the implementation of clean cooking technology and feedstock processing strategies to balance energy access and respiratory health care in the sustainable changes of bioenergy.