When clustering geocoded data, heading (i.e. the angle of travel relative to 0º North) is often a relevant variable. However, this being a circular metric, where the scale's extreme values of 0 and 360 degrees are actually the exact same direction, problems for classical clustering algorithms are created around this scale break point. Algorithms will wrongly interpret adjacent headings on either side of the 0/360 degree mark as maximally distant, when in fact they are minimally distant.
The way I get around this is to decompose heading angles into a sine and a cosine, whose combined outputs ensure measuring continuity for a complete cycle. But this has drawbacks that I'd rather avoid: It adds one more variable in need of clustering, plus my math knowledge doesn't go far enough to allow me to ensurethat the phased senoidal outputs of sine and cosine are clustered at constant rates for all angles (as opposed to, say, spiking around diagonals where sines and cosines vary at similar rates).
So here's my ask: how about having JMP appropriately cluster angular data without the need for intermediary transformations?
Thanks!