Choice of segments is common with various machine learning algorithms when dealing with training data. Is there any rationale for this choice for example say you are working on SVM classification
Hi, It depends on the data set. If you have large number of instances (pattern) then 10-fold is good to start with. But if you have few instances the you should apply LOO (leave one out) cross validation.
Hi, the choice of folds is compared by Ron Kohavi, A study of cross-validation and bootstrap for accuracy estimation and model selection, Proceedings of the 14th international joint conference on Artificial intelligence, p.1137-1143, August 20-25, 1995, Montreal, Quebec, Canada.
I've done an OCR project within a team of 4 persons based on KNN method. And we have validated our parameters using cross validation. But we haven't any idea about the size of the folds.
So we have done an experience to see relationship between percentage of folds and our parameters and we have found that with from 15% to 20% we get the best ones.