A first study on the noise impact in classes for Fuzzy Rule Based Classification Systems

The presence of noise is common in any real data set and may adversely affect the accuracy, construction time and complexity of the classifiers. Models built by Fuzzy Rule Based Classification Systems are recognised for their interpretability, but traditionally these methods have not considered the presence of noise in the data, so it would be interesting to quantify its effect on them. The aim of this contribution is to study the behavior and robustness of Fuzzy Rule Based Classification Systems in presence of noise. In order to do this, 69 synthetic data sets have been created from 23 data sets from the UCI repository. Different levels of noise have been introduced artificially in the class in order to analyze the FRBCSs when noise is present. The methods of Chi et al. and PDFC have been considered as a case study, analyzing the accuracy of the models created. From the results obtained, it is possible to deduce that Fuzzy Rule Based Classification Systems have a good tolerance to class noise.