The alarming environmental concern of air pollution has a severe global impact. Accurate forecasting can help minimize its hazardous implications well in time. Air Quality forecasting is a complex problem in the domain of time series data forecasting. In this paper we propose a novel customized air quality forecaster developed using Gated Recurrent Unit network-based Encoder-Decoder model (GRUED) of Deep Learning using Inverse Distance Weighted Interpolation for forecasting air pollutant concentrations of Delhi, India. The unique composition and customization of our air quality forecaster is a more efficient and better state of the art model for pollutant concentration prediction than its counterparts. Experimental results are indicative that the proposed model outperforms the conventional Deep Learning models. The proposed model was made to forecast air pollutant concentrations of SO2, CO, NO2 and O3. Each pollutant forecast was evaluated by computing MAE and RMSE metrices. MAE values for SO2, CO, NO2 and O3 forecasts were 60.63%, 26.83%, 33.2% and 31.33% lesser for our GRUED model as compared to conventional LSTM model. RMSE values for SO2, CO, NO2 and O3 forecasts were 43.4%, 19.5%, 26.4% and 27.7% lesser for our GRUED model in comparison to LSTM model. The effectiveness and optimal performance of the suggested approach has been established experimentally.