Improving raw readings from ozone low cost sensors using artificial intelligence for air quality monitoring

Abstract. Ground level ozone (O3) is a highly oxidising gas with very reactive properties, harmful at high levels, generated by complex photochemical reactions when "primary" pollutants from combustion of fossil materials react with sunlight. Thus, its concentration serves as an indicator of the activity of other air pollutants and plays a key role in Air Quality monitoring systems in smart cities. To increase its spatial sampling resolution over the city map, ozone low cost sensors are an interesting alternative, but they have a lack of accuracy. In this context, artificial intelligence techniques, in particular ensemble machine learning methods, can improve the raw readings from these sensors taking into account additional environmental information. In this paper, we analyse, propose and compare different techniques, reducing the estimation error in around 94 %, achieving the best results using the Gradient Boosting algorithm and outperforming the related work using sensor approximately 10 times less expensive.