The storage of great amount of data is nearly useless unless efficient computational methods are provided to analyze the data. Symbolic supervised learning algorithms are capable of generating set of knowledge rules, i.e. classifiers, to explain the data. From this set of rules is not always possible to extract rules that represent novel knowledge to the domain specialist. In this work, we propose a system based on evolutionary algorithms, designed for constructing individual knowledge rules with specific properties. We propose a representation for knowledge rules which has shown to be appropriated in the context of evolutionary algorithms. Based on that representation we also introduce rule recombination and evaluation methods implemented in a consistent way with the evolutionary paradigm. Finally, we present preliminary experimental results related to the system’s adequability.
|