An efficient correlation based adaptive LASSO regression method for air quality index prediction

Abstract

One of the adverse effects of population growth and urbanization in developing countries is air pollution. Due to which more than 4.2 million deaths occur every year. Therefore, prediction of air quality is a subject worth in-depth research and has received substantial interest in the recent years from academic units and the government. Feature selection methods are applied before prediction to identify potentially significant predictors based on exploratory data analysis. In this research work, a feature selection method based on Least Absolute Selection and Shrinkage Operator (LASSO) named Correlation based Adaptive LASSO (CbAL) Regression method has been proposed for predicting the air quality. For the experimental evaluation, cross regional data, including the concentration of pollutants and the meteorological factors of Delhi and its surrounding cities, has been taken from the Central Pollution Control Board (CPCB) Website. Further, to validate this feature selection method, various machine learning techniques have been taken into consideration and some preventive measures have been suggested to enhance the air quality. Feature selection analysis reveals that carbon monoxide, sulphur dioxide, nitrogen dioxide and Ozone are the most important factors for forecasting the air quality and the pollutants found in the cities of Noida and Gurugram have a more substantial impact on the Air Quality Index of Delhi than other surrounding cities. The model evaluation depicts that the feature subset extracted by the proposed method performs better than the complete dataset and the subset extracted by LASSO Regression with an average classification accuracy of 78%. The findings of this study can help to identify important contributors of AQI so that viable measures to improve the air quality of Delhi can be carried out.