You are talking about a "best fit" for a given data set. The most "scientific" way to calculate this "best fit" curve is based on the Least Squares Method (L.S.M.). Probably the best way to get acquainted with this method is to go to Google and do a search for either "best fit" or "Least Squares Method".
I can tell you this, though. Any predicton is just a guess! If you are happy with a guess, then the L.S.M. will help you to get a recognized result. But, remember, it is still just a "guess".
For instance, I had a friend who wanted to get compass readings in his boat and actual directions, and do a L.S.M. fitting to get a curve. I worked on this in QuickBASIC, came up with a beautiful program that lets you see all the data on an x-y graph, as well as any curve you choose, be it a straight line, a 2nd degree curve, a 3rd degree curve, etc., up to an nth degree curve, where n = number of data, up to 24, all based on the best fit curve, as predicted by the L.S.M. Well, he wasn't happy at all, because the only curve that passed through all points was the nth degree curve, and it was a very jumpy cure indeed, one that gave you no real predictablity at all!
In fact, a hand-drawn curve usually does a better fit than a L.S.M. fit, in my experience.
If you want to pursue this further, do e-mail me at: