Prediction of ambient air pollution with regression approach using machine learning

Abstract

Accurate air quality prediction is vital for managing pollution and protecting public health. This study evaluates the performance of three machine learning models – Decision Tree Regression (DTR), Linear Regression (LR), and Random Forest Regression (RFR) – to forecast the Air Quality Index (AQI) based on pollutants like SO₂, CO, O₃, NO₂, PM_2.5, and PM₁₀. Using data from the Central Pollution Control Board (CPCB), the research identifies RFR as the most reliable model. RFR achieved the highest accuracy scores: 0.872 (SO₂), 0.82 (O₃), 0.71 (NO₂), 0.91 (PM_2.5), and 0.82 (PM₁₀), outperforming DTR and LR. RFR’s ensemble learning approach effectively captures complex patterns and minimizes overfitting. The study highlights that RFR-based forecasting can support real-time pollution monitoring in Punjab. Integrating such models into environmental policies can lead to better-informed and timely decisions. This research offers a strong comparative framework and is a significant step toward developing AI-based air pollution warning systems for regional applications.