Embedding evolutionary multiobjective optimization into fuzzy linguistic combination method for fuzzy rule-based classifier ensembles

In a preceding contribution, we proposed a novel combination method by means of a fuzzy linguistic rule-based classification system. The fuzzy linguistic combination method was based on a genetic fuzzy system in order to learn its parameters from data. By doing so the resulting classifier ensemble was able to show a hierarchical structure and the operation of the latter component was transparent to the user. In addition, for the specific case of fuzzy classifier ensembles, the new approach allowed fuzzy classifiers to deal with high dimensional classification problems avoiding the curse of dimensionality. However, this approach strongly depended on one parameter defining the complexity of the final classifier ensemble and in consequence affecting the final accuracy. To avoid this tedious problem, we propose to automatically derive this parameter. For this purpose, we use the most common evolutionary multiobjective algorithm, namely NSGA-II, in order to optimize two criteria, complexity and accuracy. We carry out comprehensive experiments considering 20 UCI datasets with different dimensionality, showing the good performance of the proposed approach.