Study on accurate perception for enterprise financial risk based on stacking ensemble learning

To address the issues of high Matthews correlation coefficient, high Brier scores and low accuracy of perception in traditional methods, an accurate perception method for enterprise financial risk based on Stacking ensemble learning is proposed.Using the improved SMOTE to obtain a balanced data set of enterprise financial data, and based on the Boruta algorithm to screen the features related to financial risk in the balanced data set, using stacking ensemble learning technology and combining the prediction results of four base models to train a meta model, the obtained features are input into the trained meta model to obtain risk prediction results.Based on the prediction results, financial risks are classified into five levels to achieve enterprise financial risk perception.The experimental results show that the Matthews correlation coefficient of this method has never been lower than 0.90, the minimum Brier score is only 0.071 and the maximum perception accuracy is 98.36%.