Article: Machine learning modelling and algorithm optimisation for identifying financial risks in listed companies Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.66 pp.79 - 102 Abstract: Identifying financial risks in listed companies is crucial for capital market stability, yet traditional statistical models and single machine learning approaches struggle to address the challenges of high-dimensional nonlinearity and class imbalance in financial data. This paper proposes an ensemble learning framework integrating extreme gradient boosting, light gradient boosting machine, and random forest, while introducing an adaptive optimal threshold algorithm based on Bayesian optimisation to dynamically refine classification boundaries. Empirical analysis using a-share and global environmental, social and governance data (covering 4,837 listed companies) from 2015 to 2023 demonstrates that the ensemble model achieves an area under the curve of 0.964, surpassing the best single model by 3.5%. Adaptive optimal threshold algorithm enhances the ranking metric normalised discounted cumulative gain @10 by 12.7%, significantly improving the identification of scarce risk samples. This study provides regulators and investors with a high-precision risk screening tool. Inderscience Publishers - linking academia, business and industry through research

Title: Machine learning modelling and algorithm optimisation for identifying financial risks in listed companies

Authors: Fei Gao

Addresses: School of Accounting and Finance, The Open University of Shaanxi, Xi'an, 710061, China

Abstract: Identifying financial risks in listed companies is crucial for capital market stability, yet traditional statistical models and single machine learning approaches struggle to address the challenges of high-dimensional nonlinearity and class imbalance in financial data. This paper proposes an ensemble learning framework integrating extreme gradient boosting, light gradient boosting machine, and random forest, while introducing an adaptive optimal threshold algorithm based on Bayesian optimisation to dynamically refine classification boundaries. Empirical analysis using a-share and global environmental, social and governance data (covering 4,837 listed companies) from 2015 to 2023 demonstrates that the ensemble model achieves an area under the curve of 0.964, surpassing the best single model by 3.5%. Adaptive optimal threshold algorithm enhances the ranking metric normalised discounted cumulative gain @10 by 12.7%, significantly improving the identification of scarce risk samples. This study provides regulators and investors with a high-precision risk screening tool.

Keywords: financial risk identification; ensemble learning; adaptive thresholding; listed companies.

DOI: 10.1504/IJICT.2026.154126

International Journal of Information and Communication Technology, 2026 Vol.27 No.66, pp.79 - 102

Received: 27 Feb 2026
Accepted: 10 Apr 2026
Published online: 13 Jun 2026 *

Title: Machine learning modelling and algorithm optimisation for identifying financial risks in listed companies

Keep up-to-date