| Personal credit evaluation has always been the focus of research in the financial field.With the development of the times and technological progress,the amount of data in the credit datasets has become larger and larger,and the characteristics displayed by the datasets have become more complex.The performance of models is subject to data characteristics.There are differences in data characteristics between different personal credit datasets,It is particularly important whether the model and the dataset match.Under the background of the rapidly expansion of personal credit and the diversification of credit policies,flexibly selecting appropriate evaluation scheme according to the characteristics of credit datasets to prevent and control personal credit risks is conducive to promoting high-quality economic development and maintaining the stability of the financial system.It can be found through research that some studies treat data characteristics and model selection separately,ignore the connection between data characteristics and model selection,and use trial calculation or model selection based on experience for evaluation,making the selection and application of model lack objective basis.In view of this,based on the idea of data-characteristic-driven modeling,combined with specific credit scenarios,this article studies from two aspects of data characteristics and models,establishes the connection between data characteristics and models,and makes objective selection of models under the guidance of data characteristics.Selecting a model that matches the characteristics and verify the degree of adaption of the adaptive model.Firstly,the article explores and identifies data characteristics,The identification results show that the dataset have characteristics,such as unbalanced sample distribution,high-dimensional,and data missing;targeted data preprocessing is carried out according to the results of data characteristics identification.Then,the preprocessed dataset is visually analyzed,and it is found that it has data characteristics such as nonlinearity and data noise.Organically combines the results of data characteristics recognition and model research,and the Catboost prediction model that best matches the dataset is objectively selected according to the characteristics of the dataset itself.Finally,In order to verify the effect of the adapted model,the classification results of the Catboost model are compared with the single and other combined prediction model commonly used in the field of personal credit evaluation through experiments.The comparison results show that within the scope of a single model,the classification effect of the adapted Catboost model is better than the currently commonly used single classification model;in comparison with the combined model,the classification performance of the Catboost model is better than or similar to the complex ensemble model.Because of the simplest and prediction time is short,the Catboost model becomes the best adaptive model for the dataset.The experimental results show that the idea of data-characteristicdriven modeling is scientific and feasible in the field of personal credit evaluation.Among many models,the adapted model that under the guidance of data characteristics often has the best performance.On the premise of considering the characteristics of the data,the classification performance of the model does not depend on the complexity and sophistication.In some scenarios,the classification performance of a single model may be better than that of the combined prediction model,It cannot be blindly thought that combined prediction model is always better than single model.Personal credit evaluation research should be combined with the data characteristics of credit datasets to conduct targeted analysis,pay attention to the relationship between data characteristics and the degree of adaptation for models,select the most adaptive model rather than the general optimal model to analyze specific credit scenarios and datasets. |
PDF Full Text Request