How to determine the best training examples from a dataset for NN training?

Hi!

Neural network classification accuracy is strongly dependent on the choice of training samples. k-fold cross-validation and then averaging does not seem a good option to me. It's sometimes like teaching a kid simple things and taking an exam on the difficult problems. How can one decide the best training samples for training an NN?

Thanks in advance.

Answer 1

[Greg Heath]

There is no standard approach. One of many approaches:

 1. Standardize (zscore or mapstd)
 2. Remove or modify outliers
 3. Obtain multiple training set only designs via newrb and/or patternnet with 'dividetrain' 
 4. Remove points with more than M number of misclassifications.

Hope this helps

Thank you for formally accepting my answer

Greg

[TS Sharma]

Thank you Greg!