Research On The Early Warning Model Of Financial Distress Of Listed Companies Based On Machine Learning

This is the best era and the worst.In the era of economic globalization,every company is facing unprecedented opportunities and challenges.Every company may encounter financial difficulties due to external or internal fac-tors.Looking for a suitable model to predict the company's financial crisis can help the company be in a competitive position and provide decision support to the company's stakeholders.Therefore,it is of great practical significance to predict whether a company is facing financial difficulties.In recent decades,machine learning algorithms have developed rapidly and become an important part of contemporary methodologies.And the prediction method of financial distress of listed companies has gradually transitioned from the initial statis-tical method to the machine learning method.This paper mainly uses four machine learning algorithms commonly used in financial distress prediction of listed companies:support vector machine al-gorithm,logistic regression algorithm,decision tree algorithm,neural network algorithm,and two integrated algorithms:Stacking model and SVM-Logistic based on parallel model.This article focuses on the parameter optimization and comparative analysis of four simple machine learning algorithms,and then compares the classification effects of the two integrated algorithms and the four base models,as well as the comparative analysis of the prediction performance between the two integrated models.:The prediction performance of the four base models is the support vector machine model;the prediction effect of the two integrated models is better than the single base model;the prediction effect of the Stacking model is slightly better than the SVM-Logistic model.The main innovations of this article are:(1)In terms of data selection,the data selected in this article are company data selected based on financial information from 2017 to 2019.(2)In terms of sample selection,this paper adopts the undersampling method not used in the previous literature-the N-earMiss algorithm based on the neighboring idea to deal with unbalanced data,rather than random undersampling.(3)In terms of model selection,the inte-grated model selected in this paper is the Stacking model,which can integrate different types of base models,which is different from the integrated algorithm that can only integrate the same type of base models.