Approach to coding the ensemble structure of classifiers for solving the problem of searching for an optimal structure by means of a genetic algorithm

Kuznetsov Vladimir Vyacheslavovich

Building an effective multi-class classifier for creating two-dimensional maps of the terrain is a difficult task, one approach to which is to build an ensemble of classifiers, each classifier of which will solve a small binary classification problem. Thus, the task of compiling and training a classifier is reduced to the problem of choosing a binary classifier and training an ensemble of binary classifiers. This article discusses an approach to constructing an ensemble of classifiers based on a genetic algorithm, a classifier learning algorithm, coding the structure of the classifier ensemble. Genetic algorithms are the most common and most studied evolutionary algorithms. The result of the genetic algorithm will be the identified pattern, presented in the form of a vector. For the task of choosing a classifier structure, the result will be - the hierarchical classifier structure. The proposed approach to coding the structure of an ensemble of classifiers is necessary for encoding into a bit sequence to solve the problem of finding the optimal structure using a genetic algorithm. A comparative study of algorithms using a genetic algorithm to build an ensemble of classifiers with algorithms that do not use a genetic algorithm has been carried out. The result of the experiments is a significant reduction in the percentage of error and recognition time by classifiers constructed using a genetic algorithm.

Keywords:genetic algorithm, classifier ensemble, structure coding