I have some raw data from the pressure sensor. So that X= raw data and Y= pressure. Now that I have the experimental data, I want to plot a (calibration) curve and derive an equation of that curve so as to get a predictive behavior. My question is this, I have lots of curves-fit with very good correlations and very small RMSE. How to select which equation is most appropriate fit for it?
If it is a calibration curve, the best is using a linear or cuadratic fit.
In general - search parsimony, that is use as few parameters as possible.
If you want the most "appropriate fit", I would agree with Kjetil. Select the most parsimonious explanation (i.e., the simplest explanation appropriate to explain the relationship or variation).
If you want to include all Data that you have collected, you may wish to use Newton's or Spline interpolating Polynomials.
I agree with Kjetil and Kyle but would add that you should analyse the residuals of your candidate models and choose the one that has the most random distribution of residuals. R^2 and RMSE alone don't give you this.
- Badges
- Science topic