A Hybrid CNN-LSTM Model for Stock Price Prediction with Spatial and Temporal Dependencies

Abstract

Stock price prediction is crucial in financial decision-making and investment strategies, significantly influencing investors' profitability and market stability. This paper aims to systematically review and evaluate Machine Learning (ML) and Deep Learning (DL) methodologies, primarily focusing on Long Short-Term Memory (LSTM) networks and Convolutional Neural Networks (CNNs) for stock price forecasting. A hybrid CNN-LSTM model is proposed to enhance predictive accuracy. Specifically, the CNN component initially extracts essential spatial features from historical financial data, while the LSTM network subsequently captures the temporal dependencies to produce accurate predictions of future stock prices. Additionally, the review briefly addresses other ML techniques, including Autoregressive Integrated Moving Average (ARIMA) Model and various hybrid approaches, highlighting their strengths and limitations. This study covers diverse market conditions on publicly available financial datasets. Results indicate that the hybrid CNN-LSTM model outperforms individual LSTM and CNN models, effectively capturing both rapid fluctuations and long-term price trends. Experimental results demonstrate the proposed hybrid model's superior accuracy and potential as an effective predictive tool for investors and traders in financial markets.