Air Quality Forecasting Using Transformer and LSTM Models: A Comparative Study on Single-City Data

Biju, Alfin (2025) Air Quality Forecasting Using Transformer and LSTM Models: A Comparative Study on Single-City Data. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (2MB) | Preview
Preview	PDF (Configuration Manual) Download (299kB) | Preview

Abstract

Air quality forecasts are essential in safeguarding the health of urban citizens, as well as informing sustainable urban design. In this work, statistical and deep learning models are assessed and compared in terms of their ability to predict current and short-term pollutant concentration in Houston, Texas, based on ten-year hourly environmental data provided by the U.S. Environmental Protection Agency. The statistical benchmark was a conventional SARIMAX model, and Long Short-Term Memory (LSTM) networks and a deep learning approach were represented by a Transformer-based Informer model.

Preprocessing followed a rigorous cleaning of the dataset through cleaning, time-based feature engineering, and also generating lag variables prior to training the
models to predict pollutants like PM 2. 5, NO 2, O 3, and NO 2. The findings indicate that SARIMAX with its linear and stationary prescriptions was not able to capture peaks of pollutants as well as periodic trends. The LSTM model was superior, and it took the trends in the season and sudden change well. Nonetheless, the model with the greatest accuracy was the Informer, showing a great capacity to estimate long-range dependencies.

The results validate this assertion as deep learning models especially, those which incorporate Transformer architecture are most suitable in predicting urban air quality as compared to classical statistical approaches. This research paper advises the city dwellings to embrace the modern forecasting systems, consider benchmarking their models based on the existing state of the art Transformer models, and use explainable AI to increase transparency and stake\ti at trust.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Singh, Jaswinder UNSPECIFIED
Uncontrolled Keywords:	Air Quality Forecasting; LSTM, SARIMAX; Transformer(Informer); Deep Learning; Urban Pollution Prediction
Subjects:	G Geography. Anthropology. Recreation > GE Environmental Sciences H Social Sciences > HT Communities. Classes. Races > Urban Sociology > City Planning Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	30 Jun 2026 17:35
Last Modified:	30 Jun 2026 17:35
URI:	https://norma.ncirl.ie/id/eprint/9414

Actions (login required)

View Item