基于多特征量分析XGBoost-EMD-LSTM模型的股票价格预测

Stock Price Prediction Based on Multi-Feature Analysis Using the XGBoost-EMD-LSTM

摘要: 高效且准确地预测股票价格一直是研究者们不断追求的目标，由于股票价格呈现出明显的非线性、非平稳性及时序性的特征，提升股价预测的准确性成为了研究的重点。提出一种基于多特征量分析极端梯度提升(eXtreme gradient boosting, XGBoost)模型以及基于经验模态分解(Empirical Mode Decomposition, EMD)和长短期记忆(long short-term memory, LSTM)模型的组合预测方法对股票收盘价进行预测。首先，建立XGBoost模型对股票收盘价进行预测，并将预测结果作为新变量添加到原始数据的特征中，形成新的多元时间序列的数据；其次，用EMD算法分解新数据得到有限个本征函数；最后，建立LSTM模型并建模，得到最后的预测结果。通过仿真实验结果表明，本文的模型相比于RNN模型、单特征LSTM模型、多特征LSTM模型、XGBoost模型、EMD-LSTM模型、EMD-RNN模型，所提出模型的预测精度更高、效果更好。说明提出的方法能够有效提高预测准确性，具有一定的参考价值，对于投资者的决策具有积极的指导意义。

Abstract: Efficient and accurate stock price prediction has been a continuous pursuit for researchers. Due to the significant nonlinearity, non-stationarity, and time-series characteristics of stock prices, improving prediction accuracy has become a key research focus. This paper proposes a combined forecasting method for stock closing prices based on multi-feature analysis using the eXtreme Gradient Boosting (XGBoost) model and a hybrid model incorporating Empirical Mode Decomposition (EMD) and Long Short-Term Memory (LSTM). First, an XGBoost model is constructed to predict stock closing prices, and the prediction results are added as new variables to the original dataset, forming a new multivariate time series. Next, the EMD algorithm is applied to decompose the new data into a finite number of intrinsic mode functions. Finally, an LSTM model is built and trained to obtain the final prediction results. Stimulation results demonstrate that the proposed model outperforms the RNN model, single-feature LSTM model, multi-feature LSTM model, XGBoost model, EMD-LSTM model, and EMD-RNN model in terms of prediction accuracy and overall effectiveness. This indicates that the proposed method can significantly enhance prediction accuracy, providing valuable reference insights and offering positive guidance for investors’ decision-making.