Anomaly detection plays a very significant role in terms of the detection of adverse pollution events that can affect human health. Our study focuses on a systematic literature review, taxonomy development, and comparative analysis of different anomaly detection techniques. This study evaluates the performances of various machine learning techniques based on detecting anomalies in air quality datasets, including Long Short-Term Memory auto-encoders (LSTM - AE), clustering algorithms like K-means, and statistical approaches. Thus, this research study demonstrates the potential of different anomaly detection models besides addressing the impact of ground truth selection, and focusing on exploring hybrid techniques for model generalization and reliability.