I am using markov chain process for prediction. I created a transition matrix using training dataset .Now I want to experiment in test dataset and want to compute the accuracy of transition matrix for test dataset. How can I measure error rate of the prediction.
Mean squared errors are the solution.
At each step of your new data collection try to estimate the new state from the old one with your estimated transition matrix. When your guess is wrong, you introduce an error as the difference between your estimation and the reality. If I rember correctly your previous post, you've got your states from clustering. So you have a way to quantify error (ie distance between estimated state and real state).
Mean and square are here to make sure you can compare different models.