Dynamic Financial Distress Prediction Based On Optimized AdaBoost-SVM Model

With the gradual development and improvement of China's market economy,more and more listed companies have fallen into financial difficulties due to poor management in a highly competitive market environment.This will not only expose companies to crisis,but also affect investors and creditors.If this happens to a large number of companies at the same time,and the bad loan ratio of investors and creditors rises,it may have an impact on the entire financial system and trigger a new financial crisis.Financial distress is a comprehensive manifestation of the emergence of crises during the development of an enterprise.Since its birth,its research has been highly concerned by the industry and academia due to the importance of the problem itself.Building an accurate and effective financial distress forecasting model not only can improve the company its own level of risk management,but also can provide decision-making base for the company's managers.A review of relevant literature shows that the existing financial distress forecasting models are mainly static predictions and cannot be updated dynamically over time.In addition,when constructing a financial distress forecasting model,a balanced data matching method is tended to be used,which is inconsistent with the imbalance between the financially normal companies and the financially distressed companies in reality.Therefore,in order to make up for the shortcomings of the existing research,this paper starts from the perspective of realizing dynamic financial distress forecasting,considering the concept drift problem in dynamic forecasting and the sample imbalance problem in real situations.By introducing time weight,in order to improve the original AdaBoost algorithm to build an optimized AdaBoost-SVM model.The main work of the thesis is as follows:Firstly,by combing domestic and foreign research literature,this article introduces in detail the concept of financial distress,the method of predicting financial distress,the phenomenon of concept drift,and the imbalance in the prediction of financial distress.Based on this,combined with the research object characteristics of this article,to make a targeted definition of the concept of financial distress;at the same time,using economics and financial management theories,to analyze the causes of financial distress for enterprises in detail,and to analyze the concept of drift in the financial distress forecast,so as to construct a dynamic financial distress forecasting model lays the theoretical foundation.Secondly,this paper introduces the method of time weighting to deal with the problem of concept drift in dynamic financial distress,and assigns different weights to samples at different times.The data that is closer to the current time has a larger weight,and the data that is farther away has a smaller weight.The time weight parameter is used to adjust the influence of time weight to differentiate the importance of different time samples.Thirdly,based on the research of AdaBoost algorithm,this paper proposes to distinguish different types of samples in the process of weight initialization and update of AdaBoost algorithm,thereby solving the problem of unbalanced data.At the same time,we integrate the processing of concept drift into the weight initialization process of the AdaBoost algorithm,and based on the SVM algorithm that has advantages on small samples as the base classifier,we propose an optimized AdaBoost-SVM model.Then,the performance of the model is analyzed,and it is proved that the model achieves the transformation of the optimization target during the training process,which lays a solid foundation for improving the performance of the model.Finally,based on the data of China's A-share listed companies,an empirical study is made on the optimized AdaBoost-SVM model.In order to comprehensively measure the classification performance of the model constructed in this paper under imbalanced data,the evaluation indexes such as AUC,F value and G value are selected in this paper.The empirical results show that the model proposed in this paper is better than other traditional models in terms of the overall performance of the model and the processing capacity of unbalanced samples after statistical testing,and is more suitable for dynamic financial distress prediction.The optimized AdaBoost-SVM model constructed in this paper not only solves the problem of conceptual drift and data imbalance in dynamic financial distress,but also transforms the optimization goals of model training,and enriches the research field of dynamic financial distress prediction.Supplement;at the same time,it puts forward targeted policy suggestions to enterprises,investors,creditors and regulators,which has important theoretical and practical significance.