LSTM MODEL ENHANCED BY KOLMOGOROV-ARNOLD NETWORK: IMPROVING STOCK PRICE PREDICTION ACCURACY
Keywords:
LSTM, Kolmogorov-Arnold Network, Stock price prediction, Time series analysis, Nonlinear modelsAbstract
This study addresses the accuracy limitations of traditional LSTM models in stock price prediction by proposing an innovative hybrid model, the LSTM-KAN model. Combining the classical Long Short-Term Memory (LSTM) network with the Kolmogorov-Arnold Network (KAN), this model aims to enhance the performance of the LSTM model in predicting complex financial time series by leveraging the highly nonlinear expressive power of KAN. Through empirical analysis of historical stock data, a comparative study is conducted to examine the differences between the LSTM-KAN model and the basic LSTM model in terms of prediction error, stability, and generalization capability. The results demonstrate that the LSTM-KAN model significantly reduces prediction errors in most cases, improving prediction accuracy and providing new perspectives and tools for stock market analysis.References
[1] Chen S, Ge L. Exploring the attention mechanism in LSTM-based Hong Kong stock price movement prediction. Quantitative Finance, 2019, 19(9): 1507-1515.
[2] Kim T, Cho S. Predicting residential energy consumption using CNN-LSTM neural networks. Energy, 2019, 18272-81.
[3] Ziming Liu , Yixuan Wang , Sachin Vaidya , Fabian Ruehle , James Halverson , Marin Solja?i? , Thomas Y. Hou , Max Tegmark . KAN: Kolmogorov-Arnold Networks. arXiv preprint arXiv: 2024. 19756.
[4] Wenchao G, Zhigang L, Chuang G, et al. Stock price forecasting based on improved time convolution network. Computational Intelligence, 2022, 38(4): 1474-1491.
[5] Zhang, Zhiping, Wang, et al. Design of financial big data audit model based on artificial neural network. International Journal of System Assurance Engineering and Management, 2021, (prepublish): 1-10.