Heterogeneous ensemble algorithm for classification of different types of data

In this article developed heterogeneous ensemble algorithm for classification of different types of data is proposed. The algorithm is based on the iterative use of single (basic) classifiers on the initial training sample and inclusion in the ensemble only those classifiers whose relative error does not exceed a predetermined threshold. With the algorithm a few ensembles were designed for data from machine learning database and for real medical data. The comparative testing shows the advantages of the proposed ensemble algorithm compared with the single classifiers (the increase of classification accuracy, the decrease of the variance of the classifier).