Machine Learning-Based Forecasting of PM2.5 Concentration in Dhaka, Bangladesh Using Meteorological Data (2017–2019)

This poster presents a machine learning-based approach to forecast PM2.5 air pollution levels in Dhaka, Bangladesh, using meteorological data collected from 2017 to 2019. The study employed a Linear Regression model trained on hourly data featuring temperature, relative humidity, wind speed, solar radiation, and rainfall. After preprocessing and cleaning the dataset, the model achieved an R² of 0.43 and RMSE of 44.24 µg/m³. The findings reveal seasonal patterns and inverse correlations between PM2.5 concentrations and meteorological variables—particularly during the winter months when pollution levels peak due to stagnant atmospheric conditions. While simple in design, the model demonstrates that even basic ML tools can provide meaningful insights for air quality forecasting in low-resource settings. This work was presented at the Symposium on Information Science & Engineering (SISE), Autumn-2025, hosted by the University of Information Technology and Sciences (UITS), Dhaka, Bangladesh. Future improvements may involve nonlinear models (e.g., XGBoost, LSTM), additional features like traffic or industrial data, and integration of satellite-based inputs for more accurate and robust forecasting.