| Based on existing studies of personal identification methods, a Support Vector Machine(SVM)-based algorithm is proposed in this thesis to appraise personal credit, with an adequate loan given to the right person.Firstly, the advantages of SVM over other methods and models are presented with a proposal that application of SVM to personal credit has some advantages over existing methods. SVM has a firm support of mathematics and, thus, it is fairly precise to classify the data based on the established model.Secondly, it is important to set up a correct evaluation system of personal credit. However, many domestic publications directly borrow the same evaluation system abroad without validation. In the thesis, the differences of the evaluation systems at home and abroad are analyzed and a suitable evaluation system is set up regarding the personal credit risk situation in China.Thirdly, as to the data collected in this thesis, it is much necessary to quantify the qualitative variables, because they are critical to setting up an SVM model. In the thesis, a cross table analysis method is first used to delete and merge some characteristic variables as required; then, based on"Information Value", 12 datasets of 394 loan borrowers'data from a commercial bank are handled, with qualitative variables quantified and 7 variables from 12 datasets chosen for SVM.Finally, the data are divided into two subsets, one of which is for training and the other of which is for validation of the classification. Two different SVMs are used to do both the training and validation, respectively, with each SVM adopting four basic Kernel functions. A comparison of the two cases is made and the appropriate model is obtained. |
