Intelligent analysis of heterogeneous data based on hybrid neural network methods and adaptive normalization

Currently, there is a rapid growth in the types and volumes of accumulated data in various subject areas, which increases the need to develop effective approaches in terms of information visualization, transformation and analysis. Especially relevant is the use of computer methods, including machine learning algorithms and artificial intelligence elements, which significantly improve the accuracy and speed of information processing. Modern methods of integrating the stages of data collection, preprocessing, visualization and analysis aimed at solving applied problems in such areas as ecology, finance, medicine and personnel management are considered. A new VisNormNet method is proposed that combines visualization, normalization, and hybrid neural network processing for the interpretable analysis of diverse types of data. Unlike existing solutions, the method provides for an adaptive choice of preprocessing procedures based on a correlation structure to increase the model's resilience to outliers and distribution bias. An approach combining statistical methods, visual analysis tools and machine learning, in particular neural networks, is proposed. This approach allows for automated analysis of large and diverse datasets. Computational experiments have been conducted on open and internal samples, demonstrating the advantages of the proposed method in comparison with traditional algorithms, including logistic regression and the method of support vectors. The results showed a significant increase in the accuracy of classi-fication and forecasting, especially when using hybrid models, and can be used to build intelligent decision support systems in conditions of uncertainty and data incompleteness. The practical significance of the research lies in the development of a universal tool for specialists in the field of data analysis, focused on improving the effectiveness of decisions. The prospects of the work are related to the adaptation of the method for streaming data and its implementation in new subject areas such as industrial IoT and predictive analytics.