Нейросетевой анализ сложноорганизованных экологических данных

Discusses complicated experimental data classification and analysis using a neural network training algorithm developed by the authors. This algorithm allows combining data clustering and classification (and/or predicting) methods within a single neural network. As a result, a new class of neural networks appears that combines the alternative methods which are supervisory and nonsupervisory. In this work, it has been shown this enables to solve the problem of high vector variability in the center and at the boundaries of the same class, while vectors at boundaries of adjacent classes are highly similar. The solution of this problem that makes it hard to classify data considers high nonlinearity of hypersurfaces separating classes. To test this method, human impacts on Lake Shira (Russia) were estimated using complicated experimental data which were obtained by the authors during the long-term monitoring of antibiotic resistance in bacteria inhabiting the ecosystem.