Genetic programming of fuzzy logic production rules

John Koza (1992) demonstrated that a form of machine learning could be constructed by using the techniques of evolutionary computation with LISP statements. We describe an extension to this principle using fuzzy logic sets and operations instead of LISP. We show that genetic programming can be used to generate trees of fuzzy logic statements that optimise some external process, that these can be converted to natural language rules, and that these rules are easily comprehended by a lay audience. As an example we use financial traders. We demonstrate an application of these techniques to automating financial trading. We also show that even with minimal data preparation the technique produces rules with good out of sample performance on a range of different financial instruments.