A Low-Cost Air Quality Sensor Calibration Algorithm Using Self-Attention Network

In recent years, air pollution has escalated globally, raising awareness about the significance of air monitoring as population density continues to rise. However, the high costs associated with government air monitoring stations pose a challenge when it comes to expanding monitoring coverage across different areas. Fortunately, the advent of low-cost sensors provides a viable solution. By co-locating these sensors with government air monitoring stations, we can calibrate them to yield results that closely align with reference air quality measurements. The primary focus of this article is to leverage machine learning techniques to calibrate electrochemical sensors placed in a setup called “Airbox”. Through a comparison with meteorological station data, we achieved an $R^{2}$ value of 0.57 on the test set. These results demonstrate the effectiveness of low-cost sensors in accurately monitoring air pollution data in real-world scenarios.