Performance and population state metrics for rule-based learning systems

T. Kovacs
Proceedings of the 2002 Congress on Evolutionary Computation. CEC'02 (Cat. No.02TH8600)  
We distinguish two types of metric for the evaluation of rule-based learning systems: performance metrics are derived from the feedback to the learning agent from its teacher or environment, while population state metrics are derived from inspection of the rule base used for decision making. We propose novel population state metrics for use with learning classifier systems, evaluate them using the XCS system, and demonstrate their superiority in some cases. ¡ is customarily 50. is a form of
more » ... 0. is a form of classification accuracy metric, since it depends on the proportion of correct and incorrect classifications. B. System Error The second measure of performance in [8] is system error, which is a moving average over the last ¡ trials of the difference between the system prediction and the reward. The system prediction ¢ ¤ £ ¦ ¥ § © for action ¥ § is:
doi:10.1109/cec.2002.1004512 fatcat:rzeop3nku5cdlajc3hvwydhc3u