Machine Learning-Based Prediction of Stock Market Returns

Abstract

Focusing on the prediction of CSI 300 Index returns, this empirical study employs a hybrid CNN-LSTM-Attention model. The model integrates the strengths of CNN for local feature extraction, LSTM for temporal dependency modelling, and the Attention mechanism for key information focus, effectively capturing the multi-scale characteristics of financial data. Comparative experimental results demonstrate that multivariate models achieve superior fitting performance compared to univariate models, with the hybrid model outperforming either single model. This research validates the application value of deep learning models in financial time series forecasting, providing a novel approach for stock return prediction and offering reference for quantitative analysis and decision-making in financial markets.