A study on the application of instance selection techniques in genetic fuzzy rule-based classification systems: Accuracy-complexity trade-off

摘要: In the framework of genetic fuzzy systems, computational time required by algorithms for generating rule-based models from data increases considerably with increase number instances in training set, mainly due to fitness evaluation. Also, amount typically affects complexity resulting model: a higher generally induces generation rules. Since rules is considered one factors which affect interpretability models, large datasets bring less interpretable models. Both these problems can be tackled and partially solved reducing before applying evolutionary process. literature several instance selection have been proposed selecting without deteriorating accuracy generated The aim this paper analyze effectiveness 36 set methods when combined classification systems. Using 37 different sizes we show that some help reduce process decrease very limited their respect using overall set.