Abstract
Air quality monitoring and prediction are important for effective public health strategies, as air pollution is a major contributor to mortality. The goal is to design and evaluate different deep learning models: Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), Bidirectional LSTM (Bi-LSTM), and a hybrid Conv1D-LSTM on different input sets of features. The dataset consists of pollutant and meteorological parameters from 2019 to 2024 with hourly frequency for two monitoring stations: Colaba and Deonar. The models were trained on three sets of features: pollutant-only, meteorological-only, and combined features. Model accuracy was determined using the root mean square error, coefficient of determination, mean absolute percentage error, and explained variance score. The results indicate that the combined attributes significantly enhanced the prediction quality, with the hybrid CNNLSTM performing best at Colaba and the LSTM on meteorological attributes performing best at Deonar. The Bi-LSTM exhibited consistent performance on the feature sets. These results underscore the importance of using both pollutants and meteorological information and illustrate the efficiency of sophisticated deep learning structures for location-based air quality prediction.