This is my project work done in mines, from Multi regression analysis it is found that R2 is 0.9-0.93. But by ANN I found R2 is of 0.99. How much degree of certainty is going to answer by ANN? How can I correlate ANN to real filed conditions?
R-square is somewhat nebulous. Perhaps you should examine estimated variance of prediction errors. -
Also, perhaps you have some test data for validation.
Practically R-square value 0.90-0.93 or 0.99 both are considered very high and fall under the accepted range. However, in multiple regression, number of sample and predictor might unnecessarily increase the R-square value, thus an adjusted R-square is much valuable. Also, make sure your regression model can explain the reality besides high R-square or adjusted R-square value.
I do not know neural networks, but to compare regression model performances, you can use graphical residual analyses, and save test data to avoid overfitting. Also, the estimated variance of the prediction error uses an estimated sigma which is impacted by bias, so the estimated variance of the prediction error can be a good overall (i.e., considering variance and bias) measure of accuracy. You might find something in the following of some use:
https://www.researchgate.net/publication/313925533_Comparing_Model_Performances_Graphically
- Badges
- Science topic
- Similar topics
- Biological Science
- Microbiology
- Virology
- Virus