Title: Application of LightGBM algorithm in risk control of investment industry

Authors: Zhao Guang

Addresses: Xi'an Siyuan University, School of Business, Xi'an, 710038, China

Abstract: Bond default risk has the potential to result in losses for investors, which might influence their choice of investments. The study uses the gradient boosting decision tree algorithm framework as its starting point, choosing the indicators from the four categories of macro factors, debt characteristics, financial factors, and non-financial factors. The study then further calculates the value of the information in order to screen out the final indicators, and constructs a bond default prediction model. The model is optimised by introducing genetic algorithm to get the final optimised bond default risk warning model. The results of experimental revealed that the model's accuracy was improved by 2.3% in comparison to using a single index factor, the corresponding AUC value after incorporating the study's proposed index system into the model reached 0.9992, and the standard deviation of the model in the ten-fold cross-validation reached 0.0011. Results indicated that, when compared to the pre-improvement technique, the true rate of the study's improved model was 5.4% higher and the false-positive rate was 0.52% lower. It demonstrates that the model has higher predicted accuracy in addition to superior predictive stability, which can serve as a decision-basis for risk control in the investment industry.

Keywords: LightGBM; indicator system; risk control; bond default; genetic algorithm.

DOI: 10.1504/IJCSYSE.2026.151343

International Journal of Computational Systems Engineering, 2026 Vol.10 No.1/2/3/4, pp.133 - 144

Received: 16 Oct 2023
Accepted: 18 Nov 2023
Published online: 26 Jan 2026 *

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