DEVELOPMENT OF HYBRID NEURAL NETWORK MODELS FOR FORECASTING IN INTELLIGENT SYSTEMS

Abstract

The paper investigates the problem of improving forecasting accuracy in intelligent systems by integrating hybrid neural network models. The relevance of the work is due to the need to develop effective forecasting methods that ensure the stability of results and adaptability to changing conditions. The study aims to create a hybrid neural network forecasting model that combines different artificial neural network architectures to improve the accuracy of data analysis, noise resistance, and flexibility in the face of dynamic changes.The research methods are based on mathematical modelling, analysis of modern neural network architectures and comparative evaluation of their effectiveness. Deep learning approaches with a combination of convolutional and recurrent neural networks and optimisation algorithms are used to improve performance and reduce computational costs. Based on the analysis, a hybrid forecasting model that integrates attention mechanisms and adaptive parameter tuning algorithms is proposed, providing higher accuracy in handling complex data dependencies.The study’s results confirm that the proposed model provides better generalisability than traditional approaches.It has been proved that combining convolutional and recurrent architectures with optimisation algorithms improves forecasting accuracy and helps reduce sensitivity to changes in input data. The necessity of applying compression algorithms and adaptive parameter tuning to reduce the computational load without losing performance is substantiated.The practical value of the work lies in creating recommendations for adapting the proposed hybrid model to real conditions in the financial sector, medical diagnostics, and cybersecurity systems.Prospects for further research are focused on improving the mechanisms of adaptive parameter tuning, increasing the interpretability of results and optimising computational costs, which will help expand the possibilities of using hybrid neural network models in complex analytical tasks.