Application Of Random Forests To Enterprises Credit Assessment

Enterprises credit assessment is the key step in capital business, especially loan business, of commercial banks. The rationality and reliability of credit assessment will greatly affect the achievements of a bank.The commercial banks in China need some much better assessment methods to improve their competition ability, in which, nowadays, the traditional ratio-analysis method is popularly adopted. Recently, the intelligent models, such as Artificial Neural Networks (ANN) and Support Vector Machines (SVM), have been introduced into the credit assessment domain and have already achieved some promising results.The existing researches are mainly focused on the establishment of the single classifier. But, credit assessment data are with complex distribution and with much noise, so single classifier can not gain satisfying results. In this paper, we introduce a new classifier combination algorithm------Random Forests (RF), which is rather stable and robust with noise, to research enterprise credit assessment of commercial banks.In this paper, we have briefly introduced the theory of Random Forests and studied the applications of RF on enterprises credit assessment. We have used RF to eliminate outliers based on the so-called'outlier measure', deleting some obviously oddity samples; then, we have performed feature-subset selection based on RF which can measure the importance of the features; at last, we have utilized RF to construct the assessment model.The tolerance to noise and the methods of dealing with imbalanced classification problem are also studied based on RF. Simulations have been done and the conclusions that we have got on the choice of the parameters may form some good base for the future researches. The contrastive experiments are made by constructing RF, SVM and NN models, and the results show that RF is more suitable for credit assessment. Besides, cost-sensitive classification is introduced into RF, we expect to decrease the total cost of misclassifying and increase our model's practicality.