20
Experimental Setup
§Hold-out method for training and evaluation
–Dataset divided into 9 parts
–Classifier trained on 8 parts, tested on remaining part
–Rotate through all 9 parts
§
True non-genuine
True genuine
Assigned non-genuine
Assigned genuine
(1)
. And the results are pooled to compute the final performance measure. We use these three standard measures. Recall: the percentage of true genuine tables that are classified as genuine. Precision: the percentage of classified genuine tables that are truly genuine. F: average of P and R.