Air Quality Forecasting in London Using Random Forest Regression

Abstract

This paper explores short-term forecasting of daily PM2.5 and AQI levels in London during 2024, employing Random Forest (RF) regression as the primary modeling approach. Air quality records were utilized to compile a set of predicted values consisting of the 6 pollutants recorded (PM10, NO₂, SO₂, CO, and O₃) and daily average values obtained via data pre-processing. Temporal variables such as day-of-year and weekend indicators were incorporated to account for seasonal patterns and human activity effects. The RF model, trained on multivariate inputs, demonstrates strong predictive accuracy, with results showing close alignment between predicted and observed values. Feature importance analysis reveals PM10 as the dominant predictor for PM2.5, supported by known emission and dispersion dynamics. For future work, RF’s performance is also extended to benchmark with other regression approaches such as AutoRegressive Integrated Moving Average (ARIMA), Long Short-Term Memory (LSTM) and XGBoost in terms of its differences of prediction accuracy, model explainability and computation speed. The findings contribute to data-driven urban pollution monitoring, offering practical insights for short-term forecasting and public health early warning systems.