Volume 11, Issue 4 (2-2026)
mmr 2026, 11(4): 168-180 |
Back to browse issues page
Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
Solaymani Fard O, Hashemi S Z. A Dual-Timeframe Deep Learning Framework Based on Ichimoku Cloud and Optimized CNN for Trend Prediction in the Forex Market. mmr 2026; 11 (4) :168-180
URL: http://mmr.khu.ac.ir/article-1-3473-en.html
URL: http://mmr.khu.ac.ir/article-1-3473-en.html
1- , soleimani@um.ac.ir
Abstract: (183 Views)
Introduction
As the Forex market becomes increasingly complex, accurate trend forecasting has gained critical importance for traders and researchers. Unlike most studies that focus on price prediction, this paper introduces a novel bi-timeframe framework (1-hour and 4-hour) that integrates the Ichimoku Kinko Hyo strategy with deep learning models to predict directional movements in currency pairs.
Materials and Methods
The approach employs convolutional neural networks (CNNs) and hybrid architectures (CNN-LSTM, CNN-GRU), with hyperparameters optimized using the Particle Swarm Optimization Algorithm (PSO). Models are trained on historical EURUSD data (2019--2024) from MetaTrader5 and evaluated on eight highly correlated ($pm$80%) currency pairs. Due to the limitations of regression metrics (MAE, MSE, MAPE) in trading contexts, regression outputs are used solely for 4-hour trend classification, with Accuracy and F1-score as primary performance measures.
Results and Discussion
Results show that PSO-optimized models, particularly Ichimoku-CNN-GRU-PSO (ICGP), consistently outperform standard variants, achieving the highest Accuracy (up to 80.23% on USDSGD) and F1-score across most pairs.
Conclusion
The findings confirm that Ichimoku-based features, combined with hybrid deep learning and metaheuristic optimization, significantly enhances trend forecasting reliability and generalization in volatile financial markets.
As the Forex market becomes increasingly complex, accurate trend forecasting has gained critical importance for traders and researchers. Unlike most studies that focus on price prediction, this paper introduces a novel bi-timeframe framework (1-hour and 4-hour) that integrates the Ichimoku Kinko Hyo strategy with deep learning models to predict directional movements in currency pairs.
Materials and Methods
The approach employs convolutional neural networks (CNNs) and hybrid architectures (CNN-LSTM, CNN-GRU), with hyperparameters optimized using the Particle Swarm Optimization Algorithm (PSO). Models are trained on historical EURUSD data (2019--2024) from MetaTrader5 and evaluated on eight highly correlated ($pm$80%) currency pairs. Due to the limitations of regression metrics (MAE, MSE, MAPE) in trading contexts, regression outputs are used solely for 4-hour trend classification, with Accuracy and F1-score as primary performance measures.
Results and Discussion
Results show that PSO-optimized models, particularly Ichimoku-CNN-GRU-PSO (ICGP), consistently outperform standard variants, achieving the highest Accuracy (up to 80.23% on USDSGD) and F1-score across most pairs.
Conclusion
The findings confirm that Ichimoku-based features, combined with hybrid deep learning and metaheuristic optimization, significantly enhances trend forecasting reliability and generalization in volatile financial markets.
Keywords: Forex trend forecasting, Convolutional Neural Networks, Ichimoku strategy, Metaheuristic optimization
Type of Study: Research Paper |
Subject:
Mat
Received: 2025/12/22 | Revised: 2026/02/26 | Accepted: 2025/12/30 | Published: 2026/02/26 | ePublished: 2026/02/26
Received: 2025/12/22 | Revised: 2026/02/26 | Accepted: 2025/12/30 | Published: 2026/02/26 | ePublished: 2026/02/26
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |
