Voting Ensemble Learning Model (VELM) for Harmful Gas Detection in Environmental Applications

Abstract

The rapid industrialization and technological advancements of recent decades have significantly improved various facets of human life but have also intensified air pollution through the emission of harmful gases. Accurate detection of these pollutants, including Carbon Monoxide (CO), Ammonia (NH₃), and Sulfur Dioxide (SO₂), is critical for mitigating environmental and health hazards. Traditional sensor technologies often fail due to calibration issues and susceptibility to extreme temperatures, underscoring the need for advanced monitoring solutions. This study investigates the integration of artificial intelligence (AI), particularly ensemble machine learning models, to address these challenges. Leveraging the Gas Sensor Array Drift Dataset, a Voting Ensemble Learning Model (VELM) was developed and benchmarked against individual machine learning classifiers such as Random Forest, Support Vector Machine, and Logistic Regression. The VELM demonstrated superior performance, achieving a classification accuracy of 99.46%, surpassing conventional methods while maintaining low variance. Despite marginal accuracy differences with Random Forest, VELM’s majority voting approach consistently ensured robust performance. The findings highlight the transformative potential of ensemble learning in environmental monitoring and provide a foundation for future research. Recommendations include exploring deep learning enhancements and deploying the model in real-world settings to refine its applicability for detecting a broader range of harmful gases, thereby advancing public safety and environmental sustainability.