Fine-Tuned Machine Learning Models: Enhancing Air Quality Index Prediction Accuracy Through Hyperparameter Optimization

In fast developing areas, urban air pollution is a great problem to the human health monitoring and environmental control. This paper creates a predictive model of estimating the air quality deterioration in the high-density residential-commercial area of southwest Delhi known as Dwarka, provided by ensemble machine learning methods. This study uses three predictive tools-Linear Regression (LR), Support Vector Regression (SVR) and the Random Forest Regression (RFR) that have been trained using nine air quality constituents that are measured in ambient monitoring stations. Optimization of the model was done using stratified cross-validation (partitioning of 80:20 training and testing) and systematic hyperparameter optimization using exhaustive grid search. There are three complementary measures of error used: R2, RMSE, and MAE, which are performance benchmarking. Findings indicate that the predictive ability of Random Forest Regressor (maximum R2and minimum RMSE) is better compared to those of SVR and LR methods. The study confirms the relevance of tree-based ensemble applications in forecasting urban air quality when using high-pollution urban settings, which would make possible data-driven air quality control measures to Dwarka environmental regulation system.