Thank you for the reply and help. I am competing at Kaggle playground competition to predict NY Taxi prices. I am using Keras deep learning regression model to predict the prices. I am able to bring down the MSE to around 1.5 and doesn't budge beyond that. The biggest feature that helped me is converting latitude and longitude to three-dimensional coordinates (x, y, z). I hope this helps.
Is there a good article or website which discusses good ML features for a set of GPS coordinates. The standard ones are distance, Euclidean, Manhattan and (x, y, z) representation of GPS coordinates. Just wondering if there are others that I need to think about!