<![CDATA[Introduction: Conventional statistical models often struggle to represent complex interactions among multiple air pollutants and their non-linear associations with health outcomes. To address this limitation, this study evaluates the effectiveness of ensemble learning approaches for classifying air pollution exposure levels and predicting associated health risks across heterogeneous pollutant contexts. Methods: Two publicly accessible datasets were analyzed. The first dataset comprises toxic gas exposure measurements (CH₄, CO₂, and CO) annotated with short-term physiological health effect categories, reflecting acute exposure scenarios. The second dataset is the Jakarta Air Quality dataset (2021), which includes AQI-based criteria pollutants (PM10, PM2.5, SO₂, CO, O₃, and NO₂) representing urban ambient air quality conditions. Multiple base classifiers Decision Trees, Random Forests, Naïve Bayes, k-Nearest Neighbor, Logistic Regression, Support Vector Machines, AdaBoost, and Multi-Layer Perceptrons were implemented. Data preprocessing involved cleaning, normalization, and a 70:30 training-testing split. Ensemble strategies, particularly stacking, were developed to integrate complementary classifier strengths and improve predictive reliability. Results and Discussion: The stacking ensemble consistently outperformed individual base classifiers, achieving classification accuracies of 0.9993 for the toxic gas exposure dataset and 0.9816 for the Jakarta AQI dataset. These results indicate that ensemble learning enhances robustness, mitigates misclassification risks, and adapts effectively to variations in pollutant concentration patterns across different exposure contexts. Conclusion: Ensemble learning demonstrates strong potential as a reliable computational approach for environmental health risk assessment. Its high predictive performance supports its application in air quality management, early warning systems, and evidence-based policy development aimed at mitigating health risks associated with air pollution.]]>
