加权
梯度升压
计算机科学
Boosting(机器学习)
鉴定(生物学)
支持向量机
预测建模
财务风险
财务
数据挖掘
机器学习
随机森林
业务
植物
医学
生物
放射科
作者
Xin Jin,Zhang Wen,Xiaoyi Jiang,Lean Yu,Shouyang Wang
标识
DOI:10.1016/j.dss.2023.114090
摘要
Enterprise financial risk research mostly focuses on prediction precision, but neglects the assessment, which is unfavorable. Since assessment is an important and necessary step before prediction, its accuracy could affect the effectiveness of prediction. Therefore, we designed a new three-stage decision support research framework to discuss enterprise financial risk assessment and prediction. First, considering various factors affecting enterprise financial risk, we devised an indicator screening method based on financial risk identification ability to establish the indicator system. Second, the combined weighting method is introduced to assess financial risk. Then game theory is applied to find the optimal allocation coefficient for single methods to improve the accuracy of assessment. Third, owing to highly nonlinear risk factors, the single algorithm could not get satisfactory results; the light gradient boosting machine (LightGBM) ensemble model is applied to conduct real-value prediction rather than classification. Meanwhile, OPTUNA optimization is introduced for improving LightGBM to establish OPT-LightGBM ensemble model. The empirical results based on small and medium-sized enterprises (SMEs) in China show that our method can assess enterprise financial risk more accurately than the common method based on whether the enterprise stock is marked as special treatment (ST). Further, compared with the existing prediction models, OPT-LightGBM can improve the efficiency without losing the prediction performance, and has the best overall performance.
科研通智能强力驱动
Strongly Powered by AbleSci AI