When analyzing trajectories of airplanes or ships, one often has to deal with sensor variables like "course", which indicates the direction of travel.
The course variable exists on [0,360] and is quasi continuous in the following sense. If an object has a course of 359 degrees and then tilts a bit to the right by 2 degrees, the new course value will be 1 degree. That is, changes that cross the interval limit result in values at the other end of the interval. Similarly, if the course is 5 degrees and a course change of -15 degrees is made, the new course will be 350 degrees.
My question is if there are any good ways to recast the variable such that a prediction model for example does not have to deal with these jumps?
Is it sensible to recast the variables as "course_change", e.g., [+2, -15, ...]? If so, do we loose information on the direction of travel and can one account for that?
