Feature Engineering

When an ArduPilot quadcopter is told to return home, how much battery will the trip cost? I trained neural networks, trees and a linear model on 349 simulated flights, and the biggest gains came from telling them how fast the drone really flies home.

A satellite in low orbit is slowly falling, and there is a classical formula that predicts it. I wanted to know whether the simplest model in machine learning could do better, and the answer turned out to depend entirely on what I chose to tell it.