Embedded Genetic Learning of Highly Interpretable Fuzzy Partitions
Jorge Casillas · European Society for Fuzzy Logic and Technology Conference · 2009
A new algorithm is proposed to learn fuzzy partitions with a high interpretability degree. The set of input variables, the number of linguistic terms per variable, and the type (triangular or trapezoidal) and parameters of the membership functions are learnt by means of a meta-algorithm that uses a simple learning method to generate a fuzzy rule set from the derived fuzzy partitions. Inter- pretability constrains and powerful genetic operators are considered. A multi-objective optimization approach is used to generate different interpretability-accuracy tradeoffs. The algorithm is tested in a set of real-world regression problems with successful results compared to other methods.