Air pollution is a serious environmental problem that can negatively impact human health and environmental quality. This thesis focuses on anomaly detection in air pollutant measurements: sulfur dioxide (SO2), ozone (O3), nitrogen dioxide (NO2), nitrogen oxide (NO), carbon monoxide (CO), and particulate matter PM10. The data was obtained from ARSO. Detecting anomalies in these measurements is crucial for ensuring reliable data, as anomalies can indicate sensor malfunctions or other exceptional events. In this thesis, we implemented and compared three different machine learning models: XGBoost, LSTM autoencoder, and matrix profile. Among the analyzed methods, the XGBoost model performed the best, successfully detecting the highest number of anomalies and achieving the highest evaluation metrics.
|
