Year: 2025
Type: conference-paper
Abstract: Stock price prediction is a fundamental yet challenging task in financial markets due to their highly nonlinear and unpredictable nature. In this study, we propose a new hybrid deep learning model called CEEMD-CNN-BiLSTM-AM to enhance the accuracy of forecasting stock opening prices. Our model begins by applying Complete Ensemble Empirical Mode Decomposition (CEEMD) to split the complex stock price data into several Intrinsic Mode Functions (IMFs). This step significantly reduces noise and clarifies underlying trends. Subsequently, we use a CNN-BiLSTM-AM framework to analyze each IMF separately: CNN captures spatial features, BiLSTM identifies temporal patterns, and an attention mechanism (AM) helps prioritize important features. The individual predictions from all IMFs are then combined to generate the final price forecast. Experimental results on datasets from Kweichow Moutai (600519.SS) and Sinopec (601857.SH) demonstrate that our proposed method significantly outperforms traditional models such as LSTM, CNN, GRU, Transformer, and simple baseline models. Our approach particularly excels at identifying and predicting low-frequency price trends, making it a highly effective tool for financial forecasting. The findings suggest that CEEMD-CNN-BiLSTM-AM is a promising and robust predictive framework for stock market analysis and can enhance trading strategies.
Authors: Enli Wei, Zihao Wang
Institution: Wuhan University of Technology
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: hybrid
OpenAlex ID: W4413601795