Imbalance-Oriented Study On Enterprise’s Financial Distress Prediction

Enterprise’s financial affairs, like blood, penetrates through to the enterprise’s production and running of all areas of management by capital campaign and the management of value form. If the financial situation have some problem, it will be directly related to the survival of an enterprise. Therefore, financial risk aversion is a homework that every business must face constantly. problems must be solved in time before they happen. So, it requires that enterprises determine the current financial situation according to real-time financial data, prevent the occurrence of financial distress. Consequently, a good financial distress early warning mechanism’s appearance is becoming more and more pressing.Financial distress prediction (FDP) has always been a hot spot of concern to scholars, and many machine learning methods have been introduced, which are playing a large role for the financial distress early warning. However, most studies based on the number ratio of distressed companies to normal ones is1:1to build a prediction model of financial distress, which does not match apparently the real proportion of distressed companies in all listed companies, thus overestimating the predictive power of the model for distressed companies and resulting in predictive error rate increases, even it would have serious consequences. Therefore, this article simulates the actual situation, according to the ratio of1:3samples to collect financial distressed samples and normal ones and construct imbalanced dataset for FDP research, it will make these studies more targeted and rational.Though a large number of literatures read, this article summarizes and concludes previous research results. In accordance with the research idea of finding problem, analyzing problem and solving problem; Along the research technology route of literature read, and data collection, statistics analysis, model building and empirical analysis, this article put forward new thought and new method for listed companies’ FDP.First, it expands listed companies’ FDP research based on imbalanced perspective, in accordance with the ratio of1:3to collect financial distressed samples and non-financial distressed samples, and seeing the sample set as imbalanced dataset so that using the imbalanced classification methods for listed companies’ FDP.Then, it uses classical cost sensitive support vector machine (Support Vector Machines, SVM), SMOTE-SVM and data-set-partition-based SVM-ensemble in imbalanced classification methods to build corresponding three FDP models:C-SVM, S-SVM and D-SVM, and it analyses the strengths and weaknesses of every model for FDP by horizontal and vertical comparison.Finally, for the problem that samples scale in FDP research is limited and sub-classifiers’ number in integrated classifier is too little, this article proposes a new imbalance-oriented SVM method that combines the synthetic minority over-sampling technique (SMOTE) with the Bagging ensemble learning algorithm (SB-SVM), and afterwards based on China listed companies’ imbalanced samples dataset, it makes empirical comparison analysis to the four models and single using of SVM model. The results indicate that the new SB-SVM-ensemble model’ G-mean and F-mean are the best in the five FDP models, and its overall classification effect is optimal.