Volchenko O. Complex method of construction of solving rules of likelihood trained systems of automatic recognition

The thesis is devoted to the solution of an actual problem of development of effective methods of construction for solving rules of likelihood trained systems of automatic recognition of the open type. A method of construction of the reduced weighed training sample of meta-objects which weight pays off before construction of solving rules of classification that has allowed to reduce training sample essentially is offered and proved. The new method of correct addition of the meta-objects given in sample, necessary is offered at construction of trained systems of recognition of the open type. The method of potential functions on the weighed training meta-samples that has allowed to use at construction of solving rules of value of attributes of distinguished objects and knowledge of an arrangement of objects of training sample in attribute space offered. The genetic algorithm of construction of solving rules of classification in which corrected is developed is presented in the form of pseudo-boolean functions, the way of likelihood formation of an initial population of the chromosomes, the considering weight of objects of training sample is offered, the method of calculation double fitness-function is offered, the set of genetic operators is certain and theoretically proved. The new method of construction of solving rules of likelihood trained systems of automatic recognition of the open type, in a complex including way of definition of isolation of classes, method of formation of the weighed training sample of meta-objects, genetic algorithm of construction of the solving rule, allowed to raise quality of recognition is offered, to lower time expenses for construction of a solving rule and performance of classification, to reduce volume data.