Air pollution prediction and hotspot detection using machine learning

Abstract

Air pollution is a vital issue that affects day-to-day lives. It is observed that throughout the world, there is an instant need to overcome the monster of pollution. According to statistics, most of the polluted cities in the world are in India. This poses a serious need of the hour for the Indian scientists, engineers, and authorities as a whole to fight and reduce it as much as possible. The time has come when one needs to plan their outside activities on pollution levels and air quality status. Air Quality Index (AQI) varies daily; hence it is difficult to predict future trends for the same. The current study proposed a machine learning-based model that uses sensors, past/present pollutants concentration data, and satellite data to predict air pollution in the regions in India. We emphasize the fact that other than measurable pollutants (PM10, PM2.5, NO2, etc.); meteorological data like wind, temperature, and fire are also important factors in determining pollution. The model uses Long Short-Term Memory, which is the state-of-the-art technique used for time series prediction. The model could predict the concentration of the pollutants and calculate the AQI for the areas where data was available for the near future. The Root Mean Square Error on test data is 54. The results are quite promising and future model can be made, taking this as a base model. An inexpensive prediction technique can greatly help the administration in mitigating pollution.