# Considerations When Using Lat/Long Cords on Classifier

I'm interested in using Latitude and Longitude points as features to build a classifier model, but would like to better understand if I need to be taking any precautions when using Lat/Long cords in this way.

For example I'd like to predict a certain type of event/actor (actor1) mentioned in data below. I ran them through a number of models like KNN, NB, CART, Random Forest, etc...and got decent results, but I'm not sure if I need to take any precautions (e.g. scaling, transformations, or something else) when feeding spatial coordinates into a classifier.

Any thoughts, feedback or suggestions on how to best handle this are welcome.

• One thing we have found often works with latitude and longitude is to convert it into distance and direction ($0-2\pi$) from various relevant landmarks, in this case probably larger cities in Nigeria. Thus, if events happen near Lagos, or alternatively Kano or Ibaden (for example), it doesn't require two splits of latitude and longitude to get the city, just one split on "Lagos/not Lagos" and, farther down the tree, a split on distance. – jbowman Jan 4 at 4:22