RUNGE–KUTTA DUAL ATTENTION OPTIMIZATION FOR LSTM-BASED FINANCIAL TIME-SERIES FORECASTING: STABILITY, EFFICIENCY, AND ROBUSTNESS

Read full text Summarise with VIRAT
Authors: David Opeoluwa Oyewola, Sulaiman Awwal Akinwunmi, Joel John Taura
Cited by: 0
References: 22

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This study evaluates the performance of the Runge–Kutta Dual Attention (RUN-DA) optimization framework for hyperparameter tuning in a dual-attention Long Short-Term Memory (LSTM) model for financial time-series forecasting. The experiment was conducted using historical stock price data of MRS Oil Plc covering the period 2012–2024, representing a Nigerian financial market dataset. The proposed optimizer was compared with the Genetic Algorithm (GA) and Brown Bear Optimization Algorithm (BBOA) under consistent experimental conditions. Model performance was assessed using validation Mean Squared Error (MSE) and computational efficiency. Results show that RUN-DA achieved the lowest mean fitness value of 0.1369, compared with 0.4054 for GA and 0.1924 for BBOA. Sensitivity analysis indicated that moderate learning rates and time-step values produced more stable generalization performance. The evaluation under different market regimes further showed that RUN-DA maintained relatively lower fitness values across volatility conditions, decreasing from 0.2343 in low-volatility periods to 0.0473 in high-volatility periods, while GA and BBOA recorded higher corresponding values. In terms of computational efficiency, RUN-DA converged in 1015.23 seconds, slightly faster than GA (1073.25 seconds) and substantially faster than BBOA (2147.05 seconds). These results suggest that RUN-DA provides an effective optimization approach for improving LSTM-based financial forecasting models, although further validation on additional financial assets and evaluation metrics is recommended.

Fields of Study & Subjects

Stock Market Forecasting Methods Metaheuristic Optimization Algorithms Research Hydrological Forecasting Using AI

Publication details

Journal / source
FUDMA Journal of Sciences
Volume
10
Issue
3
Pages
220–231
Publication date
8 February 2026
Language
EN