Blockchain Papers

Maryam Maatallah, Mourad Fariss, Hakima Asaidi, Mohamed Bellouki

This study proposes a framework combining Variational Mode Decomposition (VMD) with a relevance-driven selection process to reduce noise and redundancy in financial time-series forecasting. The original time series is decomposed by VMD into intrinsic mode functions (IMFs), which are then evaluated using three relevance metrics: relative energy contribution, mutual information, and Spearman's rank correlation coefficient. These metrics identify the IMFs most strongly associated with future price movements. As opposed to conventional VMD-based approaches that treat all IMFs equally, the proposed relevance-driven selection process adapts IMF selection to the statistical properties of the analyzed market, thereby improving model generalization across different volatility conditions and forecasting horizons. This study makes three main contributions: (i) developing a relevance-driven IMF selection strategy to overcome limitations of traditional VMD methods, (ii) designing a hybrid framework that integrates multiscale decomposition with nonlinear information filtering, and (iii) conducting a comprehensive empirical evaluation of the proposed models. Experiments on hourly Bitcoin (BTC)/USD data from 2018 to 2025 show that the VMD-RDIC-deep learning models achieves strong forecasting performance. The results show that the proposed relevance-driven decomposition framework improves prediction accuracy and robustness compared with traditional statistical models, including Autoregressive Integrated Moving Average (ARIMA), as well as machine learning and deep learning approaches, highlighting its suitability for complex and volatile financial markets. Received: 18 January 2026 | Revised: 13 April 2026 | Accepted: 23 July 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement The data that support the findings of this study are openly available at https://www.kaggle.com/datasets/novandraanugrah/bitcoin-historical-datasets-2018-2024. Author Contribution Statement Maryam Maatallah: Conceptualization, Methodology, Software, Data curation, Writing – original draft, Visualization. Mourad Fariss: Software, Formal analysis, Writing – original draft. Hakima Asaidi: Validation, Investigation, Writing – review & editing. Mohamed Bellouki: Resources, Writing – review & editing, Supervision, Project administration.

Dr. Mohamed Nachat, Hassan Oukhouya, Saïd El Melhaoui, Moustapha Faizi · 7 authors

Time series forecasting plays a central role in finance, energy, and public health. Classical statistical, machine learning, deep learning, and generative approaches have all been applied to forecasting tasks in these fields, but comparisons between them are usually confined to a single domain or to models from the same family, and few studies report both univariate and multivariate results under the same conditions. This paper presents a controlled cross-domain comparison of four representative paradigms: classical statistics (Seasonal Autoregressive Integrated Moving Average with Exogenous variables, SARIMAX), gradient boosting machine learning (Light Gradient Boosting Machine, LightGBM), recurrent deep learning (Recurrent Neural Network, RNN), and generative-adversarial deep learning (Conditional Generative Adversarial Network, CGAN). Each model is evaluated on three monthly datasets with contrasting characteristics: Bitcoin prices (175 observations, high volatility), U.S. energy consumption (612 observations, strong seasonality), and U.S. cardiovascular mortality (300 observations, gradual trend with pandemic shock). Both univariate and multivariate variants are tested under the same preprocessing and one-step-ahead evaluation protocols, using eight performance metrics. The CGAN reaches the lowest MAPE on energy consumption (2.88%). On Bitcoin, the multivariate LightGBM lowers the MAPE from 28.26 to 19.25%, while on cardiovascular mortality the RNN reaches 3.34% MAPE. No paradigm performs best in every domain, and the gain from exogenous variables depends on both the paradigm and the domain.