Scholarlink

<p>Stock price forecasting is a crucial part of financial market analysis, playing a key role in investors’ decisions and market stability. This paper introduces a hybrid intelligent model for stock price prediction, called AVMD-FS-BO-Self-Attention-BiLSTM. The model integrates Adaptive Variational Modal Decomposition (AVMD), Feature Selection (FS), Bayesian Optimization (BO), self-attention mechanism, and Bidirectional Long Short-Term Memory (BiLSTM). Firstly, AVMD is used to decompose stock price data, alleviating modal aliasing and extracting key features. Then, feature selection is performed by combining the XGBoost model with SHapley Additive exPlanations (SHAP), simplifying the model. Next, BiLSTM captures long-term dependencies in time-series data, while the self-attention mechanism enhances the detection of complex interrelationships. Finally, BO optimizes the model’s hyperparameters. Experimental results show that this model excels in predicting SSE 300, STAR 50, SSE 50, and CSI 500 indices. Take the SSE 300 index as an example. Compared to the traditional LSTM model, the AVMD-FS-BO-Self-Attention-BiLSTM model reduces Mean Absolute Error (MAE) by 36.66%, Mean Squared Error (MSE) by 62.21%, Root Mean Squared Error (RMSE) by 38.53%, and Mean Absolute Percentage Error (MAPE) by 35.82%, significantly lowering prediction errors and offering a more reliable and accurate solution for stock price prediction.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Intelligent Stock Price Forecasting Integrating Adaptive VMD and Feature Selection with BiLSTM

  • 期刊论文

  • 发表日期：

    2025年11月4日

查看全文
  • Jing-Wen Wang,
  • Xing-Rui Wang

摘要

Stock price forecasting is a crucial part of financial market analysis, playing a key role in investors’ decisions and market stability. This paper introduces a hybrid intelligent model for stock price prediction, called AVMD-FS-BO-Self-Attention-BiLSTM. The model integrates Adaptive Variational Modal Decomposition (AVMD), Feature Selection (FS), Bayesian Optimization (BO), self-attention mechanism, and Bidirectional Long Short-Term Memory (BiLSTM). Firstly, AVMD is used to decompose stock price data, alleviating modal aliasing and extracting key features. Then, feature selection is performed by combining the XGBoost model with SHapley Additive exPlanations (SHAP), simplifying the model. Next, BiLSTM captures long-term dependencies in time-series data, while the self-attention mechanism enhances the detection of complex interrelationships. Finally, BO optimizes the model’s hyperparameters. Experimental results show that this model excels in predicting SSE 300, STAR 50, SSE 50, and CSI 500 indices. Take the SSE 300 index as an example. Compared to the traditional LSTM model, the AVMD-FS-BO-Self-Attention-BiLSTM model reduces Mean Absolute Error (MAE) by 36.66%, Mean Squared Error (MSE) by 62.21%, Root Mean Squared Error (RMSE) by 38.53%, and Mean Absolute Percentage Error (MAPE) by 35.82%, significantly lowering prediction errors and offering a more reliable and accurate solution for stock price prediction.