Error prediction after training of NN ?

Hi

After training the neural network we have the error values for different sub-sets.

But Is this possible that from these statistic we can predict the error in our prediction?

For example after training we can say that if some one use this NN for prediction,the resultant values will have this % of error?

Regards

Error Prediction from trained NN? - ResearchGate. Available from: https://www.researchgate.net/post/Error_Prediction_from_trained_NN [accessed Jul 13, 2015].

[Greg Heath]

Insufficient information.

Answer 1

[Ghada Saleh]

Hi Afaq,

Once a neural network is created, you can check its performance by calculating its 'R' value, which can be done in MATLAB by creating a regression plot. You can refer to the following link for more information on that.

http://www.mathworks.com/help/nnet/ug/analyze-neural-network-performance-after-training.html

Neural networks are not used for stochastic processes. The performance of a neural network highly depends on the data you provide to it. Hence, finding a confidence interval for a neural network is not possible. However, you can use models like ARIMA and ANOVA to establish such a confidence interval for your model.

I hope this helps.

Ghada