Evaluating learning classifier system performance in two-choice decision tasks: an LCS metric toolkit

A "metric toolkit" to evaluate learning classifier system performance is proposed. The metrics are shown to be superior to crude accuracy in evaluating classification performance, especially for data with unequal numbers of positive and negative cases. In addition, these metrics provide information to the researcher that is not available from crude accuracy. When used appropriately, these metrics provide accurate depictions of learning classifier system performance during training and testing in supervised learning environments.