Research On Corporate Bankruptcy Risk Assessment Based On Multi-stage Ensemble Learning Model

In recent years,China’s economy has grown rapidly,and the rapid economic development process will inevitably be accompanied with the prosperity and bankruptcy of enterprises.This has put forward higher requirements for the government’s social governance capabilities,especially market supervision and management capabilities.Corporate bankruptcy risk assessment can help the government to prepare a response plan in advance,take effective measures to respond to crises in a timely manner,and effectively improve the ability of government regulatory agencies to respond to economic operation risks.This is the only way for government market supervision and management departments to move toward "smart" market supervision.In addition,establishing an effective and reliable corporate bankruptcy risk assessment model can not only improve the enterprise’s market exit mechanism,but also prevent the domino effect of corporate bankruptcy in advance,hedge the chain reaction brought by corporate bankruptcy,and maintain the financial ecology and economic development environment.The key to accurately assessing a company’s bankruptcy risk lies in its accurate classification and identification.In previous studies,many domestic and foreign researchers have proposed some representative corporate bankruptcy risk assessment models.However,these models are difficult to exert their due effects when the data samples have problems such as outlier interference and imbalanced distribution.Based on the existing research results,this study fully considers various factors that affect corporate bankruptcy and carries out research on corporate bankruptcy risk assessment based on a multi-stage ensemble model.On this basis,this study proposes two corporate bankruptcy risk assessment models.The innovations of this study are as follows:（1）To handle the problem of outliers contained in balanced data samples,a novel multi-stage ensemble model with enhanced outlier adaptation is proposed.The model consists of the following three stages: first,for the outliers contained in data samples,a new BLOF-based outlier adaptation method is proposed to improve the adaptability of the base classifier to outliers;then,to improve the efficiency of the base classifier when processing data samples containing outliers,a new dimension-reduced feature transformation method is proposed,which combines feature reconstruction and dimensionality reduction to further enhance the base classifiers’ ability to deal with abnormalities;finally,to improve the overall prediction performance of the model,a stacking-based ensemble learning method is proposed.The stacking method is used to construct an ensemble model,which improves the ability of the ensemble model to deal with outliers.The experimental results indicate the superior performance of the proposed model and prove its significance and effectiveness.（2）To handle the problem of outliers in imbalanced data samples,a novel hybrid ensemble model with voting-based outlier detection and balanced sampling is proposed.The model includes the following three stages: first,for the outliers contained in data samples,a new voting-based outlier detection is proposed,which uses outlier enhancement features generated by a variety of classic outlier detection algorithms to enhance the base classifier ability to adapt to outliers;then,for the problem of falsely high model accuracy caused by imbalanced data samples,a new bagging-based balanced sampling method is proposed,which enhances imbalanced data processing capability of the base classifiers;finally,a stacking-based ensemble modeling method is proposed.By applying the adaptive parameter tuning method to the parameter tuning of the stacked ensemble model,the performance of the ensemble model is improved,and it is effective to improve the accuracy of corporate bankruptcy risk assessment.The experimental results indicate the superior performance of the proposed model and prove its robustness and effectiveness.The research results in this study are not only the innovation and expansion of the existing results of corporate bankruptcy risk assessment,but also provide new research methods and research perspectives for the field of corporate bankruptcy risk assessment.It can also help the government market supervision department identify risks in a timely manner and take measures to deal with the crisis.Effective measures to improve the ability of government regulatory agencies to respond to economic operation risks have important theoretical and practical significance.