I tuned an svm model by giving ranges of gamma and cost's values, from which best performance was obtained as 83.8%. Now, I want to do prediction using the same best model, in order to do so I used predict command and followed by table consisting confusion matrix was made, from which accuracy was calculated as 17.4%.

I don't understand why both values differ, in spite of having 83.8% of best performance score, why accuracy I got was only 17.4%?

If anyone out there know about this kindly throw light upon the solution of the same.

Thanks in advance!