Research On Early Warning Of Enterprise Financial Crisis Based On Multi-classification Of Machine Learnin

Enterprises are an important component of the market economy,which is related to various aspects of the national economy and people’s livelihood,such as economic stability,national taxation,technological innovation,and financial development.They are an important force in promoting sustainable and healthy economic development.Increasingly fierce competition among enterprises,rapid iteration of technology,innovation in business models,and changes in industry cycles,among other factors,pose many challenges to the sustainable development of enterprises,while also placing higher requirements on their financial management.In recent years,some enterprises have occasionally encountered financial crises due to cash flow,debt,and other issues,or have exposed many risks,causing many adverse effects on society.Enterprise financial crisis early warning management is still an important content that cannot be ignored in business operations.Improving the financial crisis early warning model can help enterprises identify potential financial risks and establish more effective financial risk management mechanisms,thereby reducing the probability of enterprises falling into financial crisis.In order to verify whether the Stacking integrated machine learning model constructed in this article can improve the ability to identify crisis enterprises under multiple classifications,this article first selects 362 enterprises that were ST implemented and not ST implemented in the Chinese Stock Market from 2019 to 2020 to form a research sample,and selects 25 financial and textual indicators from the T-2 year to form a feature set.Secondly,based on the two dimensions of profitability and solvency,the K-Means method is used to complete sample clustering and obtain a sample set consisting of three new classifications: financial health,crisis,and concern.Finally,according to the Stacking ensemble learning idea,three representative machine learning algorithms,namely,random forest model,limit gradient lifting algorithm,and artificial neural network model,are selected as base learners,and logical regression is selected as meta learners to construct an ensemble machine learning model.Finally,the classification and prediction effectiveness of the model for multi category enterprises is evaluated using AUC and F values.Through empirical analysis,this article draws the following conclusions:1.In the multiclassification of enterprises based on profitability and solvency,the accuracy of the model prediction for the financial concern category is relatively low,which is related to the boundary sample points of this category and the other two categories;2.After analyzing the importance of features,it is found that the most significant feature that has the most significant impact on the prediction effect of financial crises is the operating net cash flow current liability ratio,followed by the return on equity and total asset net profit ratio.In addition,adding management discussion and analysis intonation features can improve the effectiveness of early warning models;3.In multi category classification prediction,Stacking integrated machine learning model shows better prediction effect than single machine leaning model.