AN ENHANCED MACHINE LEARNING FRAMEWORK FOR AIR QUALITY PREDICTION

Abstract

Reliable prediction of the Air Quality Index (AQI) can support public awareness, environmental monitoring, and timely pollution-management decisions. This paper presents an Enhanced Machine Learning Model (ELMM), a Random Forest Regression-based workflow that combines data preparation with pollutant-derived feature engineering for AQI estimation. The study uses a publicly available Kaggle air-quality dataset containing pollutant measurements and meteorological variables. The workflow includes duplicate and invalid-record handling, missing-value imputation, normalization, feature selection, and derived features representing particulate-matter ratios, combined particulate matter, particulate interaction, and gas-pollutant interaction. The dataset is divided into training and testing subsets. ELMM is compared with Decision Tree Regression (DTR) and standalone Random Forest Regression (RFR) using tolerance-based accuracy, Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE). In the reported experiment, the integrated ELMM pipeline performed best across the evaluated metrics. These findings describe performance on the evaluated data split and should not be treated as proof of generalization to other locations or time periods. The contribution is the integrated prediction workflow, rather than a newly invented Random Forest algorithm.