基于双向长短时记忆网络的企业弹性能力预测模型

Prediction Model of Enterprise Resilience Based on Bi-directional Long Short-term Memory Network

传统的风险管理方法专注于识别、预测和评估可能发生的潜在风险,但当企业面临突发的、不可预期的风险时,往往束手无策。因此,学术界逐渐将风险管理的视角由预测并规避风险转变为提升企业自身对风险的承受能力和从风险中恢复的能力,也就是企业的弹性能力。文中提出了基于时序特征数据的企业弹性能力预测方法,使用Bi-LSTM对时序特征数据进行双向编码,获得企业的特征表示,并通过softmax分类器得到弹性能力分类结果。模型在中国上市公司的真实数据集中进行实验,macro-F1值达到89.0%,与RF,XGBoost和LightGBM等未使用时序特征数据的模型相比有一定提升。此外,进一步探讨了企业弹性能力的多种影响因素及其重要程度,并首次将机器学习方法应用到企业弹性能力的评估预测中,为企业应对突发风险提供了理论方法指导。

Traditional risk management methods focus on identifying,predicting and assessing potential risks.However,when enterprises are exposed to uncertainty and unexpected risks,traditional methods cannot deal with those risks.Therefore,the academia gradually shifts the perspective of risk management from predicting and avoiding risks to improving the ability of enterprises to withstand and recover from risks,that is,the enterprise resilience.This paper proposes a prediction method to predict the enterprise resilience based on temporal features,which utilizes Bi-LSTM to encode the temporal features to obtain the feature representation of every enterprise,and the classification results of enterprise resilience are obtained by a softmax classifier.The proposed method is validated on the real-world datasets from listed companies in China,and the macro-F1 value reaches 89.0%,which is improved compared with those models without considering temporal features,such as RF,XGBoost and LightGBM.This paper further discusses the importance of various factors that have an influence on the enterprise resilience.In this paper,the machine learning methods are applied to the evaluation and prediction of enterprise resilience for the first time,which provides theoretical and methodological guidance for enterprises to deal with unexpected risks.