Most of the people on earth are indeed exposed to high levels of air pollution. Air pollution is a significant contributor to a country's rising mortality and morbidity rates. The effective indicator that provides information about the level of air pollution to the people is the Air Quality Index (AQI). There are seven categories that make up the AQI value. It includes PM2.5 , PM10, CO, SO2, NOX, ozone and NH3. Each category has a different level of health concern. People when exposed to these pollutants for a long time are affected with respiratory diseases like asthma, emphysema etc. Therefore, developing a reliable AQI forecasting model is crucial to protect the people from the impact of outdoor air pollution. In this paper AQI prediction model is proposed using deep learning architectures such as Long Short Term Memory (LSTM), Bidirectional Long Short Term Memory (BILSTM) and Gated Recurrent Unit (GRU) by understanding the trends in time-series air quality data. It is found that a reliable AQI prediction model can be built using GRU algorithm.