Function Evaluation Using RMSE and R2

The most important variable is the coefficient of determination R2. This measure results in the following evaluations:

The coefficient of determination, R², can be maximal 1. In this case, the function prediction fits exactly to each measured value.
If the function would simply predict the mean of the measured output for any input data, an R² of 0 would be the result. A negative R² would mean that the prediction is worse than that simple prediction.
An R² of 1 means a perfect fit, every prediction of the function is the same as the measured data. Typically, the measured data has added noise. In this case, an R² of 1 means overfitting. You should be interested in a high R² with consideration of the noise.
Keep in mind that different signals can be measured with different quality. There might be signals where an R² of 0.6 might already be a good value. In contrast, a model for a different signal can be seen as good only if the R² is above 0.99.

The absolute error RMSE must be evaluated individually:

At best, the RMSE can be as good as the experimental repeatability.
Despite a good R², the RMSE can be too low, e. g. in case of a very large variation range of the modeled variable.
Despite a small R², the RMSE can be good enough, e. g. if the modeled variable features only a minor variance over the input parameters of the function.