Implementation of Feature Engineering in Prediction of AQI in India using Machine Learning

Abstract

Prediction of Air Quality Index (AQI) is the necessity of today’s era but for the prediction, analysis of different preprocessing techniques that can be applied, needs to be considered. In this study, first of all we explored various feature engineering techniques such as Data Imputation, Scaling, Extraction, Selection, and Data Split that can be used before applying machine learning algorithm for better results. Second, we used MLR and SVR (Linear, Gaussian) to build the prediction models. Finally, we used root mean square error (RMSE), R2, Mean Squared Error (MSE) and Mean Absolute Error (MAE) to evaluate the performance of the regression models in collaboration with the feature engineering techniques. The results shows that the performance of Linear SVR is better when coupled with imputation and robust scaler (R2=0.7557834846394744) as compared to the others, the performance of Gaussian SVR is better when coupled with the imputation only as compared to the others. In case of MLR, results (R2=0.7769187383819041) are almost same in all the 4 cases and performance degraded when PCA was applied.