Investigating the Effectiveness of Hybrid Deep Learning Networks Based on Self-Organization in Forecasting the Dynamics of Financial Markets

Yuriy Zaychenko, Educational and Research Institute for Applied System Analysis, National Technical University of Ukraine “Igor Sikorsky Kyiv Polytechnic Institute”, 37-A Beresteiskyi Ave., Kyiv, 03056, Ukraine

Yevgeniy Bodyanskiy, Artificial Intelligence Department, Kharkiv National University of Radio Electronics, 14 Nauky Ave., Kharkiv, 61166, Ukraine

Helen Zaichenko, Educational and Research Institute for Applied System Analysis, National Technical University of Ukraine “Igor Sikorsky Kyiv Polytechnic Institute”, 37-A Beresteiskyi Ave., Kyiv, 03056, Ukraine

Oleksii Kuzmenko, Educational and Research Institute for Applied System Analysis, National Technical University of Ukraine “Igor Sikorsky Kyiv Polytechnic Institute”, 37-A Beresteiskyi Ave., Kyiv, 03056, Ukraine

Abstract
The article investigates the effectiveness of applying hybrid deep learning networks based on self-organization to the problem of forecasting financial market dynamics. The focus is placed on the GMDH-Neo-Fuzzy architecture, which combines the advantages of the Group Method of Data Handling (GMDH) and Neo-Fuzzy Neurons. The feasibility of using such models for the analysis of non-stationary and nonlinear time series, which are characteristic of modern financial markets, is substantiated.
The experimental study was conducted using the time series of daily closing prices of NVIDIA Corporation (NVDA) stock for the period 2024-2026. To evaluate the effectiveness of the proposed approach, a comparison of three forecasting models was performed: classical GMDH, GMDH-Neo-Fuzzy, and LSTM. The evaluation was carried out for short-term and middle-term forecasting horizons (1, 3, 5, 7, 14, 21, and 28 days) using the metrics MSE, MAPE, and training time.
The obtained results showed that the GMDH-Neo-Fuzzy model provides the best balance between forecasting accuracy and computational efficiency. In particular, it demonstrated the best performance for most forecasting horizons, especially for intervals from 3 to 28 days, significantly outperforming classical GMDH and, in most cases, LSTM. In addition, it was found that the training time of GMDH-Neo-Fuzzy is orders of magnitude lower compared to the recurrent LSTM network.
It is concluded that hybrid deep learning networks based on self-organization are a promising tool for building intelligent forecasting systems for financial time series.

Keywords
hybrid networks, deep learning, selforganization, forecasting, time series.

Cite this paper
Yuriy Zaychenko, Yevgeniy Bodyanskiy, Helen Zaichenko, Oleksii Kuzmenko, Investigating the Effectiveness of Hybrid Deep Learning Networks Based on Self-Organization in Forecasting the Dynamics of Financial Markets , SCIREA Journal of Computer. Volume 11, Issue 1, February 2026 | PP. 29-53. 10.54647/computer520478

References

[ 1 ]	Yahoo Finance. Nvidia corporation (NVDA) stock price, news, quote & history. https://finance.yahoo.com/quote/NVDA, 2026.
[ 2 ]	T. Hall et al. A survey of machine learning methods for time series prediction. Applied Sciences, 15(11):5957, 2025.
[ 3 ]	Y. Zaychenko and O. Kuzmenko. Investigation of hybrid deep learning networks and LSTM in the short term forecasting in financial sphere. In 2022 IEEE 17th International Conference on Computer Sciences and Information Technologies (CSIT), pages 6568, Lviv, Ukraine, 2022.
[ 4 ]	Y. Bodyanskiy, Y. Zaychenko, H. Zaichenko, and O. Kuzmenko. Investigation of the effectiveness of artificial neural networks of different generations in the task of forecasting in the financial sphere. System Research and Information Technologies, 1:124137, 2025.
[ 5 ]	Ye. Bodyanskiy, Yu. Zaychenko, O. Kuzmenko, and He. Zaichenko. Investigation of the effectiveness of the hybrid system of computational intelligence based on bagging and Group Method of Data Handling in the task of stock index forecasting. In Proceedings of the IV International Scientific Symposium “Intelligent Solutions” (IntSol-2025), Uzhhorod, Ukraine, 2025. 15 May.
[ 6 ]	Yuriy Zaychenko, Helen Zaichenko, and Galib Hamidov. Hybrid GMDH deep learning networks – analysis, optimization and applications in forecasting at financial sphere. System Research and Information Technologies, 1:73–86, 2022.
[ 7 ]	Takeshi Yamakawa, E. Uchino, T. Miki, and H. Kusabagi. A neo fuzzy neuron and its applications to system identification and predictions to system behavior. In Proceedings of the International Conference, pages 477–484, 1992.
[ 8 ]	Zineb Bousbaa, Javier Sanchez-Medina, and Omar Bencharef. Financial time series forecasting: A data stream miningbased system. Electronics, 12(9):2039, 2023.
[ 9 ]	Pavlos I. Zitis, Stelios M. Potirakis, and Alex Alexandridis. Forecasting forex market volatility using deep learning models and complexity measures. Journal of Risk and Financial Management, 17(12):557, 2024.
[ 10 ]	Zhiang Qiu, Clemens Kownatzki, Fabien Scalzo, and Eun Sang Cha. Historical perspectives in volatility forecasting methods with machine learning. Risks, 13(5):98, 2025.
[ 11 ]	Tolga Omay, Aysegul Corakci, and Esra Hasdemir. High persistence and nonlinear behavior in financial variables: A more powerful unit root testing in the ESTAR framework. Mathematics, 9(20):2534, 2021.
[ 12 ]	Yung-Chen Li, Hsiao-Yun Huang, Nan-Ping Yang, and Yi-Hung Kung. Stock market forecasting based on spatiotemporal deep learning. Entropy, 25(9):1326, 2023.
[ 13 ]	Ian Goodfellow, Yoshua Bengio, and Aaron Courville. Deep Learning. MIT Press, 2016.
[ 14 ]	A. G. Ivakhnenko. Heuristic selforganization in problems of engineering cybernetics. Automatica, 6(2):207219, 1970.
[ 15 ]	A. G. Ivakhnenko and I. A. Mueller. SelfOrganization of Forecasting Models. Technika, Kiev, 1985.
[ 16 ]	A. G. Ivakhnenko, G. A. Ivakhnenko, and J. A. Mueller. Selforganization of the neural networks with active neurons. Pattern Recognition and Image Analysis, 4(2):177188, 1994.
[ 17 ]	Weina Wang, Jiapeng Shao, and Huxidan Jumahong. Fuzzy inferencebased LSTM for longterm time series prediction. Scientific Reports, 13:20359, 2023.
[ 18 ]	S. Hochreiter and J. Schmidhuber. Long shortterm memory. Neural Computation, 9(8):17351780, 1997.
[ 19 ]	Song, Xiaobao & Deng, Liwei & Hao, Wang & Zhang, Yaoan & Yuxin, He & Cao, Wenming. Deep learning-based time series forecasting. Artificial Intelligence Review. 58. 2024. 10.1007/s10462-024-10989-8.
[ 20 ]	Chen, Weisi & Hussain, Walayat & Cauteruccio, Francesco & Zhang, Xu. Deep Learning for Financial Time Series Prediction: A State-of-the-Art Review of Standalone and Hybrid Models. Computer Modeling in Engineering and Sciences. 139. 187-224. 2023. 10.32604/cmes.2023.031388.
[ 21 ]	Vahid Naghashi, Mounir Boukadoum, and Abdoulaye Banire Diallo. A multiscale model for multivariate time series forecasting. Scientific Reports, 15:1565, 2025.
[ 22 ]	Hongzhe Fang, Tianhong Li, and Huiting Xian. Comparative analysis of machine/deep learning models for singlestep and multistep forecasting in river water quality time series. Water, 17(13):1866, 2025.
[ 23 ]	Suradhaniwar, Saurabh & Kar, Soumyashree & Durbha, Surya & Jagarlapudi, Adinarayana. Time Series Forecasting of Univariate Agrometeorological Data: A Comparative Performance Evaluation via One-Step and Multi-Step Ahead Forecasting Strategies. Sensors. 21. 2021. 10.3390/s21072430.
[ 24 ]	Green, Riku & Stevens, Grant & Abdallah, Zahraa & Filho, Telmo. Stratify: unifying multi-step forecasting strategies. Data Mining and Knowledge Discovery. 39. 2025. 10.1007/s10618-025-01135-1.
[ 25 ]	Eivind Strom and Odd Erik Gundersen. Performance metrics for multistep forecasting measuring win-loss, seasonal variance and forecast stability: An empirical study. Applied Intelligence, 54:10490–10515, 2024.