Multi-classification Financial Crisis Warning Based On Ensemble Algorithm

The status of listed companies in our national economy is self-evident,their financial situation attract attention of every interest subject.The current changeable economic environment urgently requires listed companies to establish a financial crisis early-warning model to avoid being eliminated by the market.Based on a sample of computer,communication and other electronic equipment manufacturing listed companies,this paper applies machine learning algorithms to build financial crisis early-warning model for this industry.After reading extensive relevant literature and materials,this paper clarifies the research ideas and constructs the framework.Firstly,this paper reviews the research on financial crisis early warning,systematically introduces the definition and classification of financial crisis,as well as the indicators and models of financial crisis early warning in previous literatures.Secondly,the principles,advantages and disadvantages of the relevant machine learning algorithms,including the single classifiers and the ensemble algorithms,which will be used in this paper are expounded.Thirdly,this paper uses the solvency and profitability indicators to crate a four quadrant matrix classify the financial position of the samples,the index system consists of financial information and non-financial information and feature selection is used for feature reduction.Then the processed data were respectively imported into the heterogeneous classifier ensemble model established by stacking and the homogeneous classifier ensemble model established by random forest,and the output results were compared and analyzed.Finally,the generalization performance of the models is tested with the data of chemical materials and the chemicals manufacturing industry.The results show that the two financial crisis early warning models have similar performance for the same data set,and both models have great performance in identifying financial crisis enterprises.In general,the heterogeneous classifier ensemble model is better than the homogeneous classifier ensemble model.For data sets of different industries,the model still has good prediction effect,the generalization performance of the model is proved.