A comparative study of deep learning approaches for stock price prediction

Stock price prediction (SPP) is a highly complex and dynamic task, heavily influenced by non-linear patterns, seasonal effects, and economic volatility. Traditional statistical methods often struggle to model these complexities effectively. Deep learning techniques have emerged as powerful tools capable of addressing these challenges due to their ability to learn and represent relationships within large datasets. However, many existing models encounter limitations in capturing both short-term fluctuations and long-term trends simultaneously. In this study, we perform a comprehensive comparison of various deep learning approaches, including Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), hybrid CNN + RNN + Attention architectures, and Transformer models, for stock price prediction system. Additionally, we introduce an enhanced Transformer-based model that integrates BiLSTM networks with the Transformer architecture to better capture temporal dependencies by considering long-term and short-term information. Through experimental evaluations on various high-tech stock market datasets, our results demonstrate that the proposed model outperforms existing methods, achieving better accuracy and robustness across various market conditions.