HypeAIR: A novel framework for real-time low-cost sensor calibration for air quality monitoring in smart cities

Abstract

While less reliable than authorized air quality stations, low-cost sensors help monitor air quality in areas overlooked by traditional devices. A calibration process in the same environment as the sensor is crucial to enhance their accuracy. Furthermore, low-cost sensors deteriorate over time, necessitating repeated calibration for sustained performance. HypeAIR is a novel open-source framework for the management of sensor calibration in real-time. It incorporates two calibration methodologies: a combination of machine learning models (Voting Regressor and Support Vector Regression) and the Long Short-Term Memory deep learning model. To evaluate the framework, three extensive experiments were conducted over a 2-year period in the city of Modena, Italy, to monitor NO, NO2, and O3 gases. Both calibration methodologies outperform the manufacturer calibration and our baseline (i.e., a variation of the Random Forest algorithm) and maintain efficiency over time. The availability of the source code facilitates customization for monitoring additional pollutants, while shared air quality datasets ensure reproducibility.