Application of the Hybrid SVM-KNN Model for Credit Scoring

Along with the increase number of users for the credit, the screening of applicants becomes very significant. If the credit of applicants is bad, the bank will obtain a great loss. Support vector machine (SVM) is one of the most popular kinds of algorithms for the new consumer´s credit approval. However, there is a disadvantage that the more close to the optimal hyper plane, the greater possibility of marking the error label of the data. In the view of the situation that the data near the optimal hyper plane may be misclassified and the probability is very high, in this paper, we employ the hybrid model SVM-KNN algorithm to improve the prediction accuracy of SVM. This way fully combines the advantages of SVM and KNN algorithms. Two group UCI datasets are chosen in our experiments. The experimental results imply that the hybrid SVM-KNN model is a promising approach for credit scoring.