Predicting PM2.5 Concentrations in Bengaluru Using Ensemble Machine Learning Models and Explainable AI Techniques

Particulate matter PM2.5, originating from various anthropogenic and natural sources such as vehicular emissions, industrial activities, and biomass burning, is a major component of air pollution. They are extremely dangerous for health and can easily penetrate the lungs, consequently impairing lung functions. The adverse effect of PM2.5 can lead to adverse lung infections. This study aims to predict the PM2.5 concentrations using Bangalore's air quality monitoring station using diverse machine learning models. Data from the monitoring station is utilised to train and compare various algorithms like KNN, Decision Trees, Random Forest, SVM, XGBoost, etc. Factors like meteorological data, traffic density, and past pollution levels are considered as features for prediction. Performance metrics like MSE, RMSE, MAE and R-squared are used to evaluate model efficiency and accuracy. The best-performing models are chosen for an ensemble approach. Moreover, various hyperparameter tuning techniques are assessed for the Random Forest Regressor, and Explainable AI methods like LIME and SHAP are used to interpret the results. The study yields optimal parameters and impressive findings with ensemble models. These findings highlight the significance of ensemble models and hyperparameter tuning in machine learning.