# How to choose a window for curve fitting and prediction?

I am using Pareto distribution to fit a serie of survival rates (with least square).

My ultimate goal is to use this fitting curve for prediction. Thus I would mainly focus on the tail of the fitting curve. In other words, I need to define the "tail" first, and in case my survival rate serie is a time serie, I'd call it "window period".

However, I've never done curve fitting before, thus have no idea what is the best choice for such a window.

Can anyone clue me how can I achieve my goal? Appologies if I'm asking something simple.

-
There are plenty of regression methods, start with linear/polynomial regression. Or try using filters like Kalman filters which are self adaptable. –  rtybase Sep 19 '13 at 19:16