Data preprocessing and extraction of logical rules based on the genetic algorithm

The possibility of joint use of genetic algorithms and methods for extraction of logical rules in data mining problems is considered. The combined data preprocessing algorithm is suggested. It consists in segmentation and selection of the most significant values of attributes and combinations of attributes using the genetic algorithm. It can be attributed to the so-called island algorithms as the number of populations is selected equal to the number of classes. Choosing such quantity of populations is due to the fact that the performance assessment of the rules accuracy and completeness is appropriate to produce for each class. It is shown that the genetic algorithm allows selection of logical rules in the multivariate data analysis with high accuracy and completeness. The presented algorithm can be used to analyze data in data warehouses to detect hidden patterns in them, on the basis of which the construction of rules and decision trees could be produced.