Air Quality Forecasting using LSTM

Forecasting air quality is vital for public health and environmental protection, offering early alerts on pollution levels. This project utilizes Long Short-Term Memory (LSTM) neural networks to predict the Air Quality Index (AQI) in Kansas City, using over a decade of real-time data from the Environmental Protection Agency (EPA). The model analyzes daily pollutant levels, temperature, and humidity to predict air quality trends for the next 7 days, aiding authorities and individuals in making informed decisions about outdoor activities and pollution management.

To enhance forecast precision, the dataset was cleaned, normalized, and examined using correlation heatmaps to identify significant pollutant interactions. Two LSTM models were created: a univariate model focusing solely on AQI trends and a multivariate model that includes meteorological factors. The multivariate model achieved an RMSE of 6.2, indicating greater accuracy in air quality forecasting. Training took 45 minutes on a GPU, optimizing temporal pattern recognition and long-term forecasting abilities.

This project highlights the potential of deep learning in environmental forecasting. Future improvements may involve real-time deployment through API integration, transfer learning for adaptation to different cities, and expanding the dataset with satellite-based air pollution data. By merging AI with environmental science, this project supports a data-driven approach to public health and pollution management, empowering communities and policymakers with precise air quality forecasts.