Research and Application of the Bayesian financial distress prediction model

Dataset used in financial distress prediction is unbalanced. The traditional machine learning method such as neural network and support vector machine is premise with the hypothesis that the class distribution is basically balanced. The classification of unbalanced dataset inclines to the relative majority samples results in the lower identification of the minority while the conventional down-sampling results in the important information loss of the majority class. A financial distress prediction model is established based on the complete dataset of the listed companies in China by expressing the profile of expert knowledge in the way of prior probability combined with the Naive Bayesian. It is proved that compared with the classic machine learning method model, the new model gets the better predictive validity.