Deep-RNN: Removing Trends and Improving Forecasting Accuracy
IEEE Access · 2026 · 14 · 118849-118862
DOI: 10.1109/access.2026.3718960 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Hongfei Xiao und 2 weiteren Autor:innen. Die Arbeit nennt 39 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: Anhui Provincial Department of Education.
Abstract (Originalsprache)
In time series forecasting, researchers commonly assess different models using metrics such as MAE, MSE, and MAD. However, they seldom analyze forecasting errors in depth, for instance, analyzing the errors by modelling or visualization, to examine whether trend characteristics persist in the errors and if so, how these characteristics are eliminated to enhance forecasting accuracy. Given this, this paper visualizes the forecasting errors of nine representative deep-learning models. The plotted curves reveal significant trends, suggesting that these errors deviate from a white-noise distribution. Consequently, drawing inspiration from ARMA modeling, this paper constructs Deep-RNN models to eliminate these trends and improve forecasting accuracy. The experimental results indicate that Deep-RNN models can effectively eliminate the trend-like structures within its forecasting errors and significantly improve the forecasting accuracy compared to original models. Further, this outcome is consistently observed across the experiments involving multiple models and datasets. In real-world simulations, this paper finds that investors following forecasts from original deep-learning models may fail to profit and could even suffer losses. In contrast, trading based on the Deep-RNN model’s forecasts often enables them to generate stable positive returns. This suggests that the Deep-RNN modeling approach makes both theoretical and practical contributions to financial time series forecasting.
Förderung
- Anhui Provincial Department of Education
Zitieren
APA
Xiao, H., Wang, C., & Chen, Z. (2026). Deep-RNN: Removing Trends and Improving Forecasting Accuracy. IEEE Access, 14, 118849-118862. https://doi.org/10.1109/access.2026.3718960
Vancouver
Xiao H, Wang C, Chen Z. Deep-RNN: Removing Trends and Improving Forecasting Accuracy. IEEE Access. 2026;14:118849-118862. doi:10.1109/access.2026.3718960
BibTeX
@article{Xiao2026,
author = {Xiao, Hongfei and Wang, Chunsheng and Chen, Zhenji},
title = {Deep-RNN: Removing Trends and Improving Forecasting Accuracy},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {118849--118862},
doi = {10.1109/access.2026.3718960}
}
EUVIMED