Research on Credit Evaluation Model for High-Dimensional Imbalanced Data

With the rapid development of credit consumption, credit evaluation model has become the key for enterprises to avoid risks and develop credit business. Real credit data often have the characteristics of high dimensionality and class imbalance. In view of the characteristics of credit data in real scenarios, this paper combines an unsupervised clustering algorithm and machine learning classification methods to establish a multi-stage credit evaluation model that can better predict real credit data. The model can effectively process high-dimensional data features, and resampling according to the data distribution to construct multiple balanced subsets. Different training subsets and parameter perturbations are used to train different base classifiers, and multiple base classifiers are combined in the ensemble model by soft voting mechanism. By studying different classification algorithms and resampling methods, this paper compares and analyzes their effects on the overall performance of the credit evaluation model. The experimental results show that the performance of the credit evaluation model proposed in this paper is better than the benchmark models on AUC, Type I error and G-mean, and it is more suitable for forecasting high-dimensional imbalanced data.