A Machine Learning Approach to Risk Asset Management Based on the BiLSTM-Attention Model

Against the backdrop of the booming development of China's financial market and the increasing participation of residents in investment, this study focuses on improving the application efficiency of LSTM neural network models in the Chinese stock market. Based on the multivariate characteristic data of 104 constituent stocks of the Shanghai and Shenzhen 300 Index from 2021 to 2023, an improved LSTM model investment portfolio was constructed, and sensitivity analysis was deeply implemented. The results showed that improved models such as BiLSTM, LSTM Attention, and BiLSTM Attention outperformed the basic multi feature LSTM model in most evaluation metrics, confirming the effectiveness of introducing bidirectional structure and attention mechanism in improving prediction performance. Specifically, the BiLSTM Attention model performs the best under most evaluation criteria, providing the best solution for stock price prediction.