Evolutionary Approach to Data Discretization for Rough Sets Theory

摘要

This article presents the LDGen method which is based on genetic algorithm. The author proposed evolutionary approach to the solution of the discretization problem for systems that induce rules on the basis of rough sets theory. The study describes details of the method with special focus on the crossing operator. The proposed approach concerns working with multidimensional samples. Thanks to application of the author's own method of for visualizing multidimensionality, i.e. so called Pipes of Samples, it was possible to visualize up to 360 dimensions, which is usually sufficient in case of problems the Rough Sets Theory deals with. Mutation and crossing methods were developed using this visualisation so that, for real numbers, it allowed to create individuals that describe one solution of the discretization. Hence the population is a set of many complete discretizations of all the attributes.