Air pollution prediction using machine learning techniques – A concept to replace existing monitoring stations with virtual monitoring stations

Air pollution in the modern world is a matter of grave concern. Due to rapid expansion in commercial social, and economic aspects, the pollutant concentrations continue to increase and disrupt human life. Also, the pollutants are taking a toll on the entire biodiversity present on the earth. Thus, monitoring the pollutant levels is of primary importance to keep the pollutants under check. Regular monitoring enables the authorities to obtain insights into the pollutant concentrations and take the appropriate measures. However, pollutant concentration monitoring is not straightforward. This requires installing monitoring stations to collect the relevant pollutant data, which comes with high installation and maintenance costs. Also, many studies show that there is always insufficient monitoring and requires broader coverage. 

In this project, an attempt has been made to model the pollutants PM2.5, PM10, and NO₂ measured by the monitoring stations in Marienplatz and Am Neckartor in Stuttgart, Germany using Machine Learning methods. Stuttgart city, in total, has six continuous monitoring stations operated by State Institute for the Environment (LUBW). Meteorological parameters, traffic data, and pollutant information from remaining monitoring stations were considered to model the pollutants. A total of five Machine learning methods, namely Ridge regressor, Support Vector regressor, Random Forest, Extra Tess Regressor, and Xtreme Gradient Boosting, were adopted. Several scenarios were explored to find a compelling set of input parameters to estimate pollutant concentrations. Further, it is shown that a similar methodology can also be applied to estimate pollutant concentrations at other locations. This procedure was tested on the monitoring station located in Karlsruhe-Nordwest, Germany. The results showed that this method could also be applied at other locations.