Uni-norm Fuzzy Pattern Trees For Evolving Classification By Imperialist Competitive Algorithm

Abstract

Fuzzy pattern trees induction was recently introduced as a novel machine learning method for classification. Roughly speaking, a pattern tree is a hierarchical, tree-like structure, whose inner nodes are marked with generalized fuzzy logical or arithmetic operators and whose leaf nodes are associated with fuzzy predicates on input attributes. Operators perform an important role in fuzzy pattern trees. These operators include arithmetic and logical operators. Unlike arithmetic operators,logical operators that were used in these trees are not parameterized. As arithmetic operators, we can choose weighted arithmetic mean and ordered weighted arithmetic mean. There are several families which contain the standard triangular norms and conorms as special cases. This way, we would implicitly select from an infinite number of operators, just like in the case of arithmetic operators. We develop this algorithm by proposing a method to using parameterized logical operators and tuning their parameters by imperialist competitive algorithm. In experimental studies, we compare our method to previous version of algorithm, showing that our method is significantly outperformsthe previous method in terms of predictive accuracy andflexibilityin operator selection.