Submissions are evaluated using accuracy — the percentage of correct predictions out of the total number of test samples.