Genetic Based Machine Learning: Merging Pittsburgh and Michigan, an Implicit Feature Selection Mechanism and a New Crossover Operator

This paper presents, for discrete data, a new crossover operator to be used together with the Natural Coding. This new operator, differently of the already existing one, beyond possessing high speed of application, explores the search space in the same way that the crossover operator used when the binary representation is adopted. Additionally, this work presents a new way of representing the mechanism of Feature Selection. This representation provides a high economy of memory, fact that supplies to the system a double genetic exploration. The system uses a hybridization of Pittsburgh and Michigan approaches. We compare our system with the C4.5 algorithm in some datasets from UCI. Results show that the proposed system is very robust and can achieve high accuracy with simple rules.