Article: A financial data forecasting and optimisation model combining LSTM and convolutional neural networks Journal: International Journal of Reasoning-based Intelligent Systems (IJRIS) 2026 Vol.18 No.17 pp.83 - 99 Abstract: In financial time series forecasting, local fluctuations and long-term dependencies often interfere with one another, making it difficult for a single model to capture both types of features simultaneously, which leads to forecasts deviating from actual dynamics. This paper proposes a hybrid architecture that integrates convolutional neural networks with long short-term memory networks, aiming to collaboratively extract multiscale local features and sequence memory to reconstruct the intrinsic evolutionary patterns of financial data. On the publicly available daily returns of the S&P 500 index, the proposed model reduces the root mean square error by 18.3% and the mean absolute error by 15.7% compared to the baseline long short-term memory, while improving directional accuracy by 12.4%, with a parameter count comparable to that of a convolutional neural network. Experiments demonstrate that this hybrid architecture effectively balances the capture of local fluctuations with the retention of global trends, providing a robust solution for financial forecasting. Inderscience Publishers - linking academia, business and industry through research

Title: A financial data forecasting and optimisation model combining LSTM and convolutional neural networks

Authors: Min Wang; Yuying Wu

Addresses: School of Economics and Foreign Languages, Chengdu Technological University, Chengdu, 611730, China ' School of Economics and Foreign Languages, Chengdu Technological University, Chengdu, 611730, China

Abstract: In financial time series forecasting, local fluctuations and long-term dependencies often interfere with one another, making it difficult for a single model to capture both types of features simultaneously, which leads to forecasts deviating from actual dynamics. This paper proposes a hybrid architecture that integrates convolutional neural networks with long short-term memory networks, aiming to collaboratively extract multiscale local features and sequence memory to reconstruct the intrinsic evolutionary patterns of financial data. On the publicly available daily returns of the S&P 500 index, the proposed model reduces the root mean square error by 18.3% and the mean absolute error by 15.7% compared to the baseline long short-term memory, while improving directional accuracy by 12.4%, with a parameter count comparable to that of a convolutional neural network. Experiments demonstrate that this hybrid architecture effectively balances the capture of local fluctuations with the retention of global trends, providing a robust solution for financial forecasting.

Keywords: financial forecasting; convolutional neural networks; CNN; long short-term memory networks; hybrid time-series models.

DOI: 10.1504/IJRIS.2026.154540

International Journal of Reasoning-based Intelligent Systems, 2026 Vol.18 No.17, pp.83 - 99

Received: 30 Apr 2026
Accepted: 29 May 2026
Published online: 02 Jul 2026 *

Title: A financial data forecasting and optimisation model combining LSTM and convolutional neural networks

Keep up-to-date