The health dangers connected with pollutants like nitrogen dioxide (NO \(_{2}\) ), air quality monitoring, modeling, and forecasting have become crucial yet difficult topics for government agencies, legislators, and citizens. Traditional monitoring techniques are changing along with technology as air quality control becomes a higher concern. Particularly, the potential of machine learning and deep learning techniques to acquire, handle, and analyze complex, multidimensional data in ways that are difficult for traditional approaches has contributed to their widespread use. In order to forecast NO \(_{2}\) concentrations, this research investigates two sophisticated graph-based deep learning models: the Spatial Graph Convolutional Network (SGCN) and the Graph Covolutional Network (GCN), utilizing a dataset of three cities’ daily weather readings and air pollution levels. Air pollution and meteorological data collected from three Delhi cities between January 2017 and December 2020 were used to test the suggested approaches. To assess the performance of the approaches, we use root mean square error (RMSE), mean squared error (MSE), and R-squared ( \(R^2\) ) as evaluation metrics.