A Novel Hybrid Model Based on CEEMDAN and Bayesian Optimized LSTM for Financial Trend Prediction

Abstract

Financial time series prediction is inherently complex due to its nonlinear, nonstationary, and highly volatile nature. This study introduces a novel CEEMDAN-BO-LSTM model within a decomposition-optimization-prediction-integration framework to address these challenges. The Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) algorithm decomposes the original series into high-frequency, medium-frequency, low-frequency, and trend components, enabling precise time window selection. Bayesian Optimization (BO) algorithm optimizes the parameters of a dual-layer Long Short-Term Memory (LSTM) network, enhancing prediction accuracy. By integrating predictions from each component, the model generates a comprehensive and reliable forecast. Experiments on 10 representative global stock indices reveal that the proposed model outperforms benchmark approaches across RMSE, MAE, MAPE, and R2 metrics. The CEEMDAN-BO-LSTM model demonstrates robustness and stability, effectively capturing market fluctuations and long-term trends, even under high volatility.