Research of self-configurating models and procedures of genetic programming for formation of decision trees in problems of the intelligent data analysis

In this work mechanisms of a self-configuration of genetic programming algorithm for the automated decision trees formation are investigated. Known, and well proved in tasks self-configurations of genetic algorithm, the Population-Level Dynamic Probabilities (PDP) and Individual-Level Dynamic Probabilities (IDP) model are considered. At the expense of a procedures community of evolutionary operators of the choice these approaches are rather just generalized on all evolutionary algorithms in general and on algorithm of genetic programming in particular. However, the specified procedures are limited in the choice of a configuration and management of the evolution course. Such ways of development of evolutionary search as restart, introduction to population of new casual individuals, cardinal change of parameters and change of search resources (addition of iterations, expansion of population, etc.) are hard to include in PDP and IDP. Besides, decision-making process, i. e. change of a configuration of search algorithm, is hidden from the user. The user can observe only results of this choice. In the offered work alternative approach to a self-configuration of evolutionary algorithms by means of the fuzzy controller is considered. Procedure of decision-making and management of a search configuration in fuzzy logical systems is similar to a reasoning of the expert and is easily generalized on the majority of ways and settings of evolutionary search which are applied in the work by the experienced user. Besides, the user can include those heuristic rules and procedures which uses in the practice in the fuzzy controller. In the work the basic possibility of application of an fuzzy control system for a self-configuration of genetic programming algorithm in a problem of the automated formation of trees of decision-making is shown. The minimum set of fuzzy rules and linguistic variables allowing operating evolutionary search is offered. Potential of the fuzzy controller and a way of increase of self-configuration procedure efficiency are discussed. Comparison of self-configuration procedures efficiency is carried out on practical tasks: classifications of irises of Fischer and forecasting of side effects at treatment of epilepsy. The analysis of the statistical importance of distinctions in efficiency of approaches is carried out, and results are discussed. The hybrid evolutionary algorithm of the automated formation of decision trees on the basis of genetic programming with the realized procedures of a self-configuration can be applied in various areas including in space-rocket branch.