Abstract
Air quality significantly impacts human health, particularly in urban areas, leading to global morbidity and mortality. Elevated air pollutant levels pose health risks, emphasizing the need for timely monitoring and detection. This study adopts an innovative approach to identify anomalies of daily NO2 concentration levels in a district of Ankara, Turkey. Leveraging both traditional statistical approaches and state-of-the-art techniques, the research aims to provide real-time alerts. Employing a multivariate strategy, the study generates new features based on historical and current data, and incorporates periodic variables, as well. Among the methods explored, Variational Autoencoder emerges as noteworthy, exhibiting superior performance with %98 recall, %82 precision and %0.12 false alarm rate. This approach not only demonstrates a high true positive rate, enhancing its efficacy in anomaly detection but also effectively mitigates false alarms, preventing alert fatigue. By using advanced methodologies with a focus on NO2 levels, the study contributes to proactive measures for public health, enabling prompt responses to potential air quality issues.