Bitcoin Price Prediction Based on ROA-VMDAlgorithm and CNN-SK-Transformer Model

Driven by its unique production, issuance mechanisms, and markettransaction characteristics, Bitcoin's price exhibits pronounced nonlinear fluctuation features, rendering prediction tasks highly complex. To address this challenge, we propose a Bitcoin price prediction model integrating RIME Optimization Algorithm (ROA)-based Variational Mode Decomposition (VMD), Convolutional Neural Network (CNN), and an improved Transformer (SK-Transformer). Firstly, ROA is employed to optimize the core parameters of VMD (number of modes \(K\) and penalty factor \(\alpha\)). The optimized VMD method is then applied to decompose the Bitcoin price series into multiple subsequences. Subsequently, these subsequences are reorganized into three sequences --- low-frequency, medium-frequency, and high-frequency --- based on their fuzzy entropy values. The low-frequency components are trained using a CNN model, while the medium-frequency and high-frequency components are modeled via the SK-Transformer architecture. Predictions from these models are aggregated to generate the final forecast, which is evaluated using multiple accuracymetrics.Experimental validation demonstrates that the ROA-VMD-CNN-SK-Transformer model outperforms alternative prediction models across all evaluation metrics,showcasing superior predictive precision.