A Novel Hybrid Framework for Stock Price Prediction Integrating Adaptive Signal Decomposition and Multi-Scale Feature Extraction

To address the issue of low prediction accuracy due to the inherent high noise and non-stationary characteristics of stock price series, this paper proposes a novel stock price prediction framework (CVASD-MDCM-Informer) that integrates adaptive signal decomposition with multi-scale feature extraction. The framework first employs a CVASD module, which is a variational mode decomposition (VMD) method adaptively optimized by a porcupine optimization (CPO) algorithm, to decompose the original stock price series into a series of intrinsic mode functions (IMFs) with different frequency characteristics, effectively separating noise and multi-frequency signals. Subsequently, the decomposed components are input into a prediction network based on Informer. In the feature extraction phase, this paper designs a multi-scale dilated convolution module (MDCM) to replace the standard convolution of the Informer, enhancing the model’s ability to capture short-term fluctuations and long-term trends by using convolution kernels with different dilation rates in parallel. Finally, the prediction results of each component are integrated to obtain the final predicted value. Experimental results on three representative industry datasets (Information Technology, Finance, and Consumer Staples) of the US S&P 500 index show that, compared to several advanced baseline models, the proposed framework demonstrates significant advantages in multiple evaluation metrics such as MAE, MSE, and RMSE. Ablation experiments further validate the effectiveness of the two core modules, CVASD and MDCM. The study indicates that the framework can effectively handle complex financial time series, providing a new solution for stock price prediction.