Title: LSTM neural network combined with data mining techniques for financial crisis early warning model construction

Authors: Jingwen Liu

Addresses: College of Economics and Management, Henan Polytechnic, Zhengzhou, 450000, China

Abstract: In enterprise financial management, early warning and prevention of financial crises are key to ensuring the long-term healthy development of the enterprise. The purpose of the research is to more effectively capture the potential factors of financial crisis and take corresponding preventive measures. Therefore, a new financial crisis warning model has been introduced in the study. This model combines the traditional K-means clustering algorithm with the improved genetic algorithm of cross model-K-means (CMGAK) for financial data pre-processing. It is further augmented with the non-negative Garrote relaxation long-short-term memory (NNG-LSTM) algorithm for comprehensive examination of financial data. The results showed that in the validation experiment of the effectiveness of the financial crisis warning model constructed by integrating the CMGAK algorithm with the NNG-LSTM algorithm, the highest accuracy of the model's prediction in the multi-step dimension was 92.54% in the two-year step sizes of T-2 to T-3. The financial crisis warning model, which combines the advantages of the CMGAK algorithm and the NNG-LSTM algorithm, has been demonstrated to perform particularly well in medium- and long-term forecasting. The model provides effective technical support for the prevention of financial crises in enterprises.

Keywords: genetic clustering; LSTM; financial crisis; early warning model; data mining.

DOI: 10.1504/IJADS.2026.153506

International Journal of Applied Decision Sciences, 2026 Vol.19 No.3, pp.289 - 313

Received: 03 Jan 2025
Accepted: 25 Mar 2025
Published online: 12 May 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article