Research Progress and Trends of Deep Learning in Stock Price Prediction: A Systematic Review from LSTM to Transformer

Stock price prediction is a key issue in quantitative finance. Due to the characteristics of the financial market, traditional prediction methods often fail to achieve accurate and stable prediction results. In recent years, deep learning technology, especially architectures such as Long Short-Term Memory (LSTM) and Transformer, have stood out with their excellent and stable performance. This paper systematically reviews the update and iteration process of stock prediction methods and elaborates on the classification of mainstream deep learning models, analysing the characteristics, advantages and limitations of different models, and comparing the differences in data sets, indicators and performance in empirical studies. In addition, it further discusses challenges such as data noise, overfitting, interpretability and computational efficiency, and thereby looks forward to future research directions such as multimodal information fusion, interpretable artificial intelligence and real-time adaptive learning. This paper aims to provide a complete technical roadmap for researchers in the field of stock price prediction to apply deep learning methods systematically.