Advancing Stock Return Prediction: A Comprehensive Study of Traditional Machine Learning and Deep Learning Models

Stock markets exhibit high volatility, driven by numerous factors, making accurate stock return prediction a challenging task. This paper provides a comprehensive review of machine learning models used to stock return forecasting, comparing traditional models like Linear Regression (LR), Support Vector Machine (SVM), and Random Forest (RF) with deep learning techniques such as Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), and Long Short-Term Memory (LSTM) networks. Traditional methods are praised for their interpretability and computational efficiency but frequently fail to convey the non-linear and time-dependent complexities of financial data. In contrast, deep learning models, particularly LSTM, are highly effective in modeling long-term dependencies in stock data but face challenges related to overfitting, interpretability, and generalizability across different market conditions. The paper also highlights the limitations of current models, including their inability to integrate external factors such as geopolitical events and policy changes. To address these issues, potential future research avenues are explored, focusing on enhancing interpretability using interpretability techniques as well as leveraging transfer learning and domain adaptation to improve scalability and model robustness. These advancements could significantly enhance the practical application of machine learning in stock return prediction, offering more reliable and interpretable solutions for real-world financial decision-making.