| Author |
 Topic  |
|
|
vflope
Brazil
2 Posts |
 Posted - 06/03/2016 : 2:27:01 PM
|
Dear all, sorry by my bad english.
1) I wondering about use of R2, R2adjusted and R2predicted. A lot of people said these estimators can NOT be used for nonlinear regressions. If it was true, why Origin provide them in nonlinear fitting tool?
2) What are the differences among RMSE, Standard Error of the Regression (S or SE) and Reduced Chi-Square?
3) What are the best estimators to compare two or more nonlinear regressions and avoid overfitting? (Assuming R2 is not feasible..)
Thank you very much!
V. Lopes |
|
|
Echo_Chu
China
Posts |
Posted - 06/06/2016 : 03:21:33 AM
|
Hi, V. Lopes
Yes, the R^2 is not a efficient quantity in evaluating the nonlinear model. but, some users still want this information...
However, if you want to compare models, I would suggest you use the Rank Models tool (accessible from menu Analysis: Fitting: Rank Models.) We also provide AIC and BIC values in the tool to evaluate the goodness of fit.
RMSE is for Standard Error of the Regression, which is the square root of Reduced Chi-Square.
Echo
OriginLab Technical Support
|
 |
|
|
vflope
Brazil
2 Posts |
Posted - 06/06/2016 : 1:27:23 PM
|
Dear Echo
Many thanks for your kind answer.
V. Lopes |
|
|
| |
Topic |
|
|
|
RMSE, S, Nonlinear fitting comparison