Feature Engineering with Latitude and Longitude




I am working on Kaggle crime category prediction problem. Here input variable of dataset are datetime, district, dayofweek, address and geo variables( Latitude and Longitude). I know feature engineering always help you to boost your model prediction power. Can you please suggest the methods to create new variable from latitude and longitude? I know, this will improve the power of model because as we go more granular compare to district it will predict the crime category more accurately.


How to handle Lat/Lon features in a model?

One thing you can do is use K-means clustering method to develop the cluster.