基于Lasso和CatBoost融合模型的信用风险评估研究

Credit Risk Assessment Based on Lasso and CatBoost Fusion Model

摘要: 准确高效的信用评估模型能够提高预测风险的能力，对信用贷款风险起到有效的防范作用。因此，我们提出了一种Lasso和CatBoost的融合模型及Lasso-CatBoost算法并将其应用于信贷数据。该算法先用Lasso进行特征选择，再用CatBoost结合重要特征进行信用风险分析，达到用较少的变量得到较高预测准确率的目的。为了验证该算法的高效性及准确性，我们以某银行信用数据为样本，在对数据进行预处理后，运用Lasso-CatBoost算法对该数据做信用风险分析。结果表明，以AUC作为评价指标，与Logistics回归算法以及RandomForest、Adaboost、XGBoost、LightGBM等集成算法相比，Lasso-CatBoost算法用到的特征最少且预测的准确率最高，说明了我们提出的算法在信用风险分析方面表现最好。

Abstract: Accurate and efficient credit evaluation model can improve the ability to predict the risk, and play an effective role in preventing credit loan risk. Therefore, we propose a fusion model of Lasso and Catboost and Lasso-Catboost algorithm and apply it to credit data. Firstly, Lasso is used for feature selection. Then CatBoost is used to analyze credit risk by combing the selected important features, which obtains higher prediction accuracy with fewer variables. In order to verify the efficiency and accuracy of the algorithm, we take the credit data of a bank as a sample. After preprocessing the data, we adopt Lasso-CatBoost algorithm to analyze the credit risk of the data. The results show that Lasso-CatBoost algorithm uses the least features and has the highest prediction accuracy, compared to Logistic regression algorithm and Randomforest, AdaBoost, Xgboost, Lightgbm and other integrated algorithms, which shows that our algorithm has the best performance in credit risk analysis.