【摘要】 针对数据不平衡导致的信用风险识别精度低的问题，利用随机欠采样算法对数据集平衡处理后，采用Logistic回归模型以及随机森林、决策树、XGboost和支持向量机等分类算法分别建立模型并进行预测。实证结果表明，随机欠采样算法可以将信用卡欺诈风险的预测精度从低于75%提升至85%以上，且G-mean和AUC等衡量非平衡数据分类性能的指标均有明显提高，该算法能够有效缓解数据不平衡导致的风险预测性能低下的问题。更多还原

【Abstract】 Aiming at the problem of low accuracy of credit risk identification caused by imbalanced data, the random under-sampling algorithm was used to balance the dataset. Then the Logistic regression model and random forest, decision tree, XGboost as well as support vector machine classification algorithms were used to establish the model and make prediction respectively. The research results show that through the random under-sampling algorithm, the prediction accuracy of credit card fraud risk is improved from less than 75% to more than 85%, and the indicators such as G-mean and AUC to measure the performance of imbalanced classification are significantly improved, which can effectively alleviate the problem of low performance of risk prediction caused by data imbalance.更多还原