Rapid technological advancement has resulted in severe air pollution in urban areas. Air quality modeling and forecasting remain vital to ensure sustainable development. Accurate regional air quality modeling necessitates the consideration of the following: (i) spatiotemporal features in air quality data and (ii) temporal features in meteorological parameters that have a large influence on the region’s air quality. However, the existing models for regional air quality modeling do not account for the above two. In this, a hybrid deep learning architecture named Spatio-Temporal Feature Fusion (STFF) is proposed for air quality forecasting. The proposed architecture is a hybrid combination of Bidirectional Long Short Term Memory (Bi-LSTM) and Spatio Temporal Graph Convolution Network (STGCN). The temporal trends in the data are extracted by Bi-LSTM and are fused with the spatiotemporal features in the air quality data extracted by STGCN for forecasting air quality. Also, the relevance of the output with respect to the identified features are interpreted using a multi-output SVR in the output layer which adds interpretability to the proposed model. The use of STGCN to extract spatiotemporal features eliminates the need for grid formation, thus, reducing pre-processing work and round off errors associated with grids. It also enables the spatiotemporal pattern extraction for monitoring stations that cannot be structured into defined grids. The proposed approach is evaluated using the real word dataset of Delhi, collected from the official website of CPCB and the results demonstrate that the STFF outperforms the state of art approaches available for air quality forecasting.