Missing value imputation for the particulate matter concentration using machine learning

Abstract

For the past few years, Delhi has been considered one of the most polluted cities in India. The people of here suffer from air pollution related diseases (Asthma, respiratory inflammation etc.) almost all the year round. The amount of air pollution is much higher for urbanization. Harmful pollutants like NO2, SO2, ground-level O3, PM2.5, PM10 are abundant in the air here. Of which PM2.5 is one of the harmful pollutants. It is very important to determine the concentration of these elements in the air to reduce pollution. Although many monitoring stations have been set up for this purpose, they often fail to provide accurate information. In this work concentration of PM2.5 has been predicted with the help of some machine learning and deep learning models. Concentration of PM2.5 has been taken at one-hour intervals for a whole year in twenty-eight cities of Delhi. In order to get the best accuracy, the models have been hyperparameter-tuned. Mean Absolute Error (MAE), Root Mean Square Error (RMSE), R2-Score metrices has been used to measure accuracy. Feature selection has also been done using different methodologies to see if better results are available. K-Nearest Neighbor (KNN) algorithm for regression has given us the best accuracy in this work