[Solved] Data splitting for SVM parameters selection vs Neural Networks

njasaj

Hi,
I have a question about SVM parameter selection and data spliting. I need to  know that is it sufficient and efficient to split the whole data set into just 2 part (train set and test set) then use cross validation on train set and select the C and gamma which to lead the best performance. Some people split data to 3 sets( train, cross-validation, test) for neural networks and select the parameters when the performance of cross-validation start to reduce and train performance increase.Is split the data set into just 2 parts and cross validate on train set good enough and acceptable for  modeling with support vector machine or the same procedure(split into 3 set) should be done for SVM? Is splitting into 2 part procedure applicable to Neural Network?

Thanks.

MariusHelf

Hi,

since the cross validation already performs the split into training and test set it should be sufficient to use *only* the cross validation and not split the data manual at all. If you are unsure, just increase the number of folds in the X-Validation.

Best regards,
Marius

njasaj

Thank you Marius.