Research on Financial Distress Prediction of Listed Companies Based on Unbalanced Data Processing and Multivariable Screening Methods

In the context of domestic supply side structural reform, the market environment is complex and ever-changing, and corporate debt defaults occur frequently. It is necessary to establish a timely and effective financial distress warning model Most of the existing distress prediction models have not effectively solved problems such as imbalanced datasets, unstable selection of key prediction indicators, and randomness in sample matching, and are not suitable for the current complex and changing market conditions in China Therefore, this article uses the Bootstrap resampling method to construct 1000 research samples, and uses LASSO (Least absolute shrinkage and selection operator) variable selection technology to screen key predictive factors to construct a logit model for predicting ahead of 3 years. In the prediction stage, the samples are randomly cut and predicted 1000 times to reduce random errors The results indicate that the Logit dilemma prediction model constructed by combining Bootstrap sample construction method with LASSO has stronger predictive ability compared to the traditional application of "similar industry asset size" method In addition, the embedded Bootstrap Lasso logit model has better predictive performance than mainstream O-Score models and ZChina Score models, with an accuracy increase of 10%, and is more suitable for China's time-varying market. The model constructed in this article can help corporate stakeholders better identify financial difficulties and make timely adjustments to reduce corporate bond default rates or avoid corporate defaults