Research on Price Prediction based on WDBiLSTM-Attention model

Gold price has the characteristics of nonlinearity, high noise and stochasticity, and accurate prediction of price trend is important for investors and international environment. The experiment selects 5-minute high-frequency data of gold futures prices from Shanghai Futures Exchange, uses Wavelet Domain Denoising (WD) to process the price series data, and uses XGBoost model in integrated learning algorithm to evaluate and filter indicators to reduce the influence of noise and some weakly correlated features on price prediction. The Wavelet Domain Denoising processed price series data and the correlated features screened by the XGBoost model are constructed as a dataset to build a hybrid model of WD- BiLSTM-Attention to predict prices. The experiments show that the WD-BiLSTM (Bi-directional Long-Short Term Memory)- Attention hybrid model improves the prediction accuracy compared with benchmark models such as Support Vector Machine (SVM), Long-Short Term Memory (LSTM) and BiLSTM and provides some reference value for the design and development of quantitative investment strategies in the future.