Evolutionary concept learning with constraints for numerical attributes

This paper proposes two alternative methods for dealing with numeri-cal attributes in inductive concept learning systems based on genetic algo-rithms. The methods use constraints for restricting the range of values of the attributes and novel stochastic operators for modifying the constraints. These operators exploit information on a subset of thresholds on numer-ical attributes. The methods are embedded into a GA based system for inductive logic programming. Results of experiments on various data sets indicate that the methods provide effective local discretization tools for GA based inductive concept learners.