In recent years, credit evaluation has become an issue of increasing concern for financial institutions. However, since most research focuses on the risk classification process, the problem of data imbalance is ignored. In real data sets, there are often far more users with good credit than users with bad credit, and the imbalance of data often easily leads to a decline in the classification performance of the model. Therefore, previous research is very limited in practical application scenarios. In this paper, we establish a new integration method for credit evaluation, which is classified into three steps: First, data preprocessing. Before inputting samples into the model, we take a series of preprocessing steps, such as missing data processing, data dimensionality reduction. Secondly, in view of the imbalance problem, the data is divided into multiple clusters using an unsupervised clustering algorithm, and the SMOTE method is used to generate minority samples in the clusters whose ratio exceeds the threshold. Finally, the GBDT2NN and Factorization Machine methods are integrated to classify the samples. In order to verify the effectiveness of this method, we use the Kaggle competition data set for verification. The results show that this method is better than other algorithms in the field of credit evaluation in terms of recall rate and AUC value.