I have a customer trying to perform K Means Clustering using data that includes (in part) degree direction (0-360) but north is problematic as those values can be near 0 or 360. Looking for suggestions for a workaround. Thank you.
Hi Peter,
You might try converting the degrees to radians by multiplying by pi()/180, then computing 2 new columns: the sine and cosine of the radian measure. I've done that for 10,000 randomly chosen points on the unit circle, and the clusters (in this case, I chose 6) seem reasonable. Notably, there is no break in clusters where sin == 0 and cos == 1. How the conversion of a single column into two affects the "distance", in the presence of several other variables, I'd need to think about a bit...
