Modeling XCS in class imbalances: Population size and parameter settings

This paper investigates the scalability of population size in XCS (eXtended Classifier Systems) to efficiently handle class imbalances in data. It develops models to predict the effect of the imbalance ratio on population initialization and classifier dynamics, demonstrating that when appropriately adjusted, XCS's population size can scale linearly with the imbalance ratio instead of exponentially. Recommendations are provided to improve XCS mechanisms for better performance in imbalanced problem domains.