Big Data Solutions in Air Pollution Forecasting

Abstract

Air pollution, especially PM10 and PM2.5, poses serious health and environmental risks worldwide, making accurate forecasting increasingly important. This study aims to develop a reliable machine learning-based system to predict PM concentrations in Budapest. To achieve this, models such as Gradient Boosting, Random Forest, Linear Regression, and deep learning models (LSTM, GRU) were applied. The data was cleaned, interpolated, and enriched with temporal and meteorological features, then evaluated using RMSE and R² metrics. Results showed that Gradient Boosting (RMSE = 4.25 μg/m³ and R² = 0.811) delivered the most accurate predictions for both short-term and day-ahead forecasts, outperforming other models. Although the study was limited by data size and lack of spatial and emission-specific factors, it provides a useful foundation for improving air quality forecasting and supporting better public health decisions.