MetaCost is a meta-algorithm which makes error-based classifiers making their decision based on the cost of errors. For example, sending advertisement is cheap, so it might be worth a lot of false positives to get a single person who is actually interested in the advertisement.

The algorithm is given in pseudocode in the paper.

Important notation:

* $C(i, j)$: Cost of predicting an example belongs to class $i$, where in fact it belongs to class $j$.