Protecting the environment from air pollution is the greatest challenges faced globally. The Environment is continuously affected by poisonous concentration in air due to human activities. As a result, quality of air is declining day by day with increase in air pollution like PM10, PM2.5, S02, N02, Ozone, Co, NH3. Machine Learning and Artificial Neural Networks are used to estimate the air quality index. Air Pollution data is available abundantly in Central Pollution Control Board (CPCB) of different cities. The paper emphasis on how the pollution data set of a metropolitan city is gathered to develop a forecasting model to predict air quality index (AQI). The approach also identifies the best fit neural network training algorithms converging at a faster rate with maximum predictive accuracy with minimum iterations. The paper also focuses on evaluating and comparing the performances with respect to, convergence time, error functions, corelation coefficient, number of epochs, and predictive accuracy of three different Neural network functions such as trainlm, trainbr and trainscg. The application of Neural network training functions on air pollution data set proposes decision making capabilities for government bodies to take necessary actions in smart city to predict AQI. The task is carried out using MATLAB tool.