The Study Of Credit Risk Model In Commercial Banks Based On SVM-KNN

With the development of economy globalization and financial liberalization, it is becoming more and more important and urgent for commercial banks to perform effective credit risk management and there are many researchers having focused on this area. Different from the traditional methods, modern credit risk management tends to apply efficient models to recognize, measure and control the risk, which always emphasize on the integration of quantitative and qualitative analysis. So in this background, this thesis puts forward a simple credit risk assessment model using the combination of Support Vector Machine and k-Nearest Neighbors.At first, we simply introduce the common theories and methods of credit risk management, including the goals, elements and various credit risk measuring methods. Then, after principles and application of Support Vector Machine based on Statistical Learning Theory and k-Nearest Neighbors presented, the combination algorithm is discussed and then inducted to establish a credit risk model, using Matlab software to realize. Based on this model, we carry out an empirical analysis to get some conclusions. In this process, Principal Components Analysis is applied to set the input attributors and some parameters selecting methods are employed. The result of the experiment shows that SVM-KNN model has a better classification performance than SVM and k-NN themselves and the percentage of train set to total set has an influence on the predict result. Besides, the probability of two types error is also discussed. Finally, we make a summary and prospect for the future study.To sum up, this thesis combines the knowledge in the area of credit risk management, machine learning technology, statistical analysis methods and financial management, considering theoretical research and empirical research at the same time. The credit risk model we established deploying Pattern Recognition method is testified to have a good effect and applicability.