Hi Sasha ... 

Thanks for the reply. 

So, I divide my initial dataset (X) into 2 parts - training [Xtr] and test [Xtst]. 

Subsequently, k-fold cross validation should divide my data [Xtr] into a secondary training set [(k-1) folds] and a validation set [1 fold]. 

After training with cross validation, the overall prediction accuracy for Xtr was always significantly higher than that of Xtst. 

Like I mentioned in my previous post, I have, however, largely resolved the issue by increasing the size of the Xtr dataset so that it is more representative of the dataset as a whole (X). 

Thanks again