Double K-Folds in SVM

Abstract

In the K-folds cross validation process for Support Vector Machine (SVM) arguments determination, checking data has taken part in the seeking of the arguments values, hence the prediction accuracy tested by checking data is not independent. To avoid this condition, double K-folds are proposed in this study. (K-1)-folds are used for data training for the best SVM arguments determination, and the Kth fold is reserved for data checking. There are 10 data sets are used to check the proposed double K-folds methods. Without doubt, the learning accuracy in K-folds is better than that in double K-folds. However, it indicated that the results of double K-folds are almost as good as those of traditional K-folds.