The effect of missing data on learning classifier system learning rate and classification performance

Missing data pose a potential threat to learning and classification in that they may compromise the ability of a system to develop robust, generalized models of the environment in which they operate. This investigation reports on the effects of the various types of missing data, present in varying densities in a group of simulated datasets, on learning classifier system performance. It was found that missing data have an adverse effect on learning classifier system (LCS) learning and classification performance, the latter of which is not seen in See5, a robust decision tree inducer. Specific adverse effects include decreased learning rate, decreased accuracy of classification of novel data on testing, increased proportions of testing cases that cannot be classified, and increased variability in these metrics. In addition, the effects are correlated with the density of missing values in a dataset, as well as the type of missing data, whether it is random and ignorable, or systematically missing and therefore non-ignorable.