I constructed a neural networks in R using neuralnet package.

I want to test that using cross-validation, that is a technique based on using 4/5 of the dataset to train the network and the fifth one as the test set.

I wonder about what measures I should use to measure the neural networks performance in terms of predictability.

Could you suggest what measures are commonly used in the field and explain me why?

Any hint and ideas about that will be appreciated.


1 Answer 1


Typical predictive performance measures used to compare accuracy of ANN models are:

  • RMSE - (root mean squared error) measures the distance between estimated and actual outcomes. Other metrics that measure the same concept are MSE, MAE or MPE.
  • R square - (R^2 or coefficient of determination) measures the reduction of variance when using the model.

When comparing two different ANN models for performance, metrics that take into account the complexity of the model may be used, such as AIC or BIC.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.