A Multi-Level LSTM-K-Means Deep Learning Framework for Robust Stock Prediction and Risk-Controlled Quantitative Investment

Abstract

In recent years, the stock market has attracted increasing attention. The inherent volatility of stock prices, often influenced by national and social policies, poses significant challenges for investors seeking profitable returns. With the rapid development of artificial intelligence, computers have demonstrated outstanding capabilities in handling complex mathematical problems. Consequently, efforts to leverage computational power to analyze and predict stock market trends have been growing. However, existing methods suffer from limited long-term sequence modeling capabilities and struggle to select candidate factors that align with individual investment preferences from a vast array of features. To address these issues, this paper proposes a deep learning factor-based comprehensive prediction model combining LSTM and K-Means. The multi-level LSTM-K-Means integrated prediction approach overcomes traditional neural networks’ shortcomings in processing long sequences and nonlinear data by incorporating stock returns and volatility to accurately identify potential high-quality stocks. Furthermore, a multi-factor scoring stock selection strategy, coupled with a fixed-percentage stop-profit and stop-loss mechanism, is designed to effectively control trading risks and enhance the robustness and profitability of quantitative investment. Experimental results demonstrate that the proposed method alleviates gradient vanishing problems, optimizes stock selection and risk management processes, and provides strong support for investors to achieve excess returns.