A machine-learning-dependent early warning system for corporate financial crises

Enterprise financial data is multidimensional and highly complex, in order to efficiently and accurately achieve enterprise financial crisis warning. Proposed and designed a machine learning based enterprise financial crisis warning system research. Firstly, principal component analysis was used to perform a dimensionality reduction on 106 financial crisis indicators of enterprises. And by improving the convolution layer of the convolutional neural network to extract the feature values of the warning indicator data after one dimensionality reduction, the extracted data features are compressed through the pooling layer to achieve a second dimensionality reduction. By utilizing a fully connected layer, the features of the data obtained through secondary dimensionality reduction are transformed into feature vectors and input into the classification layer, thereby achieving multi warning of financial crises in enterprises. Research has shown that after using PCA method to reduce the dimensionality of financial crisis indicators in experimental enterprises, the accuracy of the warning model has increased from the highest 0.83 to 0.75, and the training speed has decreased from 3 seconds to about 1 second; the average absolute error percentage of early warning is about 5.32%. It has practicality.