bsuirДоклады БГУИРDoklady BGUIR1729-76482708-0382БГУИР10.35596/1729-7648-2025-23-5-99-104bsuir-4213Research ArticleСтатьиАдаптация нейросетевого классификатора к неустойчивым входным даннымNeural Network Classifier Adaptation to Unstable Input DataМацкевичВ. В.MatskevichV. V.

Мацкевич Вадим Владимирович, канд. техн. наук, доц. каф. информационных систем управления, 

220030, Минск, просп. Независимости, 4.

Тел.: +375 29 125-49-07.

Vadim V. Matskevich, Cand. Sci. (Tech.), Associate Professor at Information Management Systems Department,

4, Nezavisimosti Ave., Minsk, 220030.

Тел.: +375 29 125-49-07.

matskevich1997@gmail.comГоЦзижаньGuoJiran

Го Цзижань, асп. каф. информационных систем управления, 

Минск.

Jiran Guo, Postgraduate at Information Management Systems Department,

Minsk.

Белорусский государственный университетBelarusian State University20252910202523599104Copyright © Мацкевич В.В., Го Ц., 20252025Мацкевич В.В., Го Ц.Matskevich V.V., Guo J.Данная работа распространяется под лицензией Creative Commons Attribution 4.0.This work is licensed under a Creative Commons Attribution 4.0 License.https://doklady.bsuir.by/jour/article/view/4213

Сегодня задачи классификации решаются, как правило, с применением нейронных сетей. В то же время при выборе архитектуры сети особое внимание уделяется сжимающим слоям, а архитектура многослойного персептрона выбирается интуитивно, хотя качество решения во многом зависит от вида разделяющей поверхности. В статье предлагается алгоритм проверки эффективности архитектуры многослойного персептрона на основе анализа свойств входных данных. Алгоритм основан на специальном, разработанном авторами, числовом показателе эффективности архитектуры нейронной сети – коэффициенте перекоса. Коэффициент рассчитывается на основе матрицы несоответствий, что не требует большого объема вычислений. Экспериментально показано, что выбор правильной архитектуры классификатора может повысить качество решения на 50 %.

Today, classification problems are typically solved using neural networks. When choosing a network architecture, particular attention is paid to compression layers, while the multilayer perceptron architecture is chosen intuitively, although the quality of the solution largely depends on the type of separating surface. This article proposes an algorithm for testing the effectiveness of a multilayer perceptron architecture based on an analysis of input data properties. The algorithm is based on a special numerical metric for assessing the effectiveness of a neural network architecture, the skew coefficient, developed by the authors. The coefficient is calculated based on the confusion matrix, which does not require a large amount of calculations. Experiments have shown that choosing the right classifier architecture can improve solution quality by 50 %.

нейронные сетимногослойный персептронархитектураклассификацияобучениевходные данныесжатиеоптимизационная задачапространство решенийметод отжигаметод градиентного спускаneural networksmultilayer perceptronarchitectureclassificationtraininginput datacompressionoptimization problemsolution spaceannealing methodgradient descent methodReferencesZhuang H., Lin Zh., Yang Y., Toh K.-A. (2025) An Analytic Formulation of Convolutional Neural Network Learning for Pattern Recognition. Information Sciences. Elsevier. 686. https://doi.org/10.1016/j.ins.2024.121317.Zhuang H., Lin Zh., Yang Y., Toh K.-A. (2025) An Analytic Formulation of Convolutional Neural Network Learning for Pattern Recognition. Information Sciences. Elsevier. 686. https://doi.org/10.1016/j.ins.2024.121317.Li Z., Yu Y., Yin Ch., Shi Y. (2025) AdaNet: A Competitive Adaptive Convolutional Neural Network for Spect ral Information Identification. Pattern Recognition. Elsevier. 163. https://doi.org/10.1016/j.patcog.2025.111472.Li Z., Yu Y., Yin Ch., Shi Y. (2025) AdaNet: A Competitive Adaptive Convolutional Neural Network for Spect ral Information Identification. Pattern Recognition. Elsevier. 163. https://doi.org/10.1016/j.patcog.2025.111472.Talebzadeh H., Talebzadeh M., Satarpour M., Jalali F., Farhadi B., Vahdatpour M. S. (2024) Enhancing Breast Cancer Diagnosis Accuracy Through Genetic Algorithm-Optimized Multilayer Perceptron. Multiscale and Multidisciplinary Modeling, Experiments and Design. 7 (4), 4433–4449. https://doi.org/10.1007/s41939024-00487-3.Talebzadeh H., Talebzadeh M., Satarpour M., Jalali F., Farhadi B., Vahdatpour M. S. (2024) Enhancing Breast Cancer Diagnosis Accuracy Through Genetic Algorithm-Optimized Multilayer Perceptron. Multiscale and Multidisciplinary Modeling, Experiments and Design. 7 (4), 4433–4449. https://doi.org/10.1007/s41939024-00487-3.Gao Y., Hu Z., Chen W. A., Liu M., Ruan Y. (2025) A Revolutionary Neural Network Architecture with Interpretability and Flexibility Based on Kolmogorov-Arnold for Solar Radiation and Temperature Forecasting. Applied Energy. Elsevier. 378 (A). https://doi.org/10.1016/j.apenergy.2024.124844.Gao Y., Hu Z., Chen W. A., Liu M., Ruan Y. (2025) A Revolutionary Neural Network Architecture with Interpretability and Flexibility Based on Kolmogorov-Arnold for Solar Radiation and Temperature Forecasting. Applied Energy. Elsevier. 378 (A). https://doi.org/10.1016/j.apenergy.2024.124844.Du J., Zeng J., Chen Ch., Ni M., Guo Ch., Zhang Sh., et al. (2025) Acoustic Emission Monitoring for Damage Diagnosis in Composite Laminates Based on Deep Learning with Attention Mechanism. Mechanical Systems and Signal Processing. Elsevier. 222. https://doi.org/10.1016/j.ymssp.2024.111770.Du J., Zeng J., Chen Ch., Ni M., Guo Ch., Zhang Sh., et al. (2025) Acoustic Emission Monitoring for Damage Diagnosis in Composite Laminates Based on Deep Learning with Attention Mechanism. Mechanical Systems and Signal Processing. Elsevier. 222. https://doi.org/10.1016/j.ymssp.2024.111770.Altalhan M., Algarni A., Alouane M. T. H. (2025) Imbalanced Data Problem in Machine Learning: A Review. IEEE Access. 13, 13686–13699. http://doi.org/10.1109/ACCESS.2025.3531662.Altalhan M., Algarni A., Alouane M. T. H. (2025) Imbalanced Data Problem in Machine Learning: A Review. IEEE Access. 13, 13686–13699. http://doi.org/10.1109/ACCESS.2025.3531662.Krasnoproshin V. V., Matskevich V. V. (2024) Random Search in Neural Networks Training. Pattern Recognition and Image Analysis. 34 (2), 309–316. http://dx.doi.org/10.1134/S105466182470010X.Krasnoproshin V. V., Matskevich V. V. (2024) Random Search in Neural Networks Training. Pattern Recognition and Image Analysis. 34 (2), 309–316. http://dx.doi.org/10.1134/S105466182470010X.Heydarian M., Doyle T. E., Samavi R. (2022) MLCM: Multi-Label Confusion Matrix. IEEE Access. 10, 19083–19095. http://dx.doi.org/10.1109/ACCESS.2022.3151048.Heydarian M., Doyle T. E., Samavi R. (2022) MLCM: Multi-Label Confusion Matrix. IEEE Access. 10, 19083–19095. http://dx.doi.org/10.1109/ACCESS.2022.3151048.Zhou X., Bu Q., Matskevich V., Nedzved A. (2024) Landscape’s Non-Natural Changes Detection System by Satellites Images Based on Local Areas. Pattern Recognition and Image Analysis. 34 (2), 365–378. http://dx.doi.org/10.1134/S1054661824700159.Zhou X., Bu Q., Matskevich V., Nedzved A. (2024) Landscape’s Non-Natural Changes Detection System by Satellites Images Based on Local Areas. Pattern Recognition and Image Analysis. 34 (2), 365–378. http://dx.doi.org/10.1134/S1054661824700159.Krasnoproshin V. V., Matskevich V. V. (2023) Neural Network Software Technology Trainable on the Random Search Principles. Open Semantic Technologies for Intelligent Systems, Research Papers Collection. 7, 133–140.Krasnoproshin V. V., Matskevich V. V. (2023) Neural Network Software Technology Trainable on the Random Search Principles. Open Semantic Technologies for Intelligent Systems, Research Papers Collection. 7, 133–140.

The authors declare that there are no conflicts of interest present.