一种面向高维缺失不平衡数据的信用评估方法  

Credit evaluation method for high dimensional missing unbalanced data

作　　者：樊东醒 叶春明[1] Fan Dongxing;Ye Chunming(School of Management,University of Shanghai for Science&Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学管理学院,上海200093

出　　处：《计算机应用研究》2021年第9期2667-2672,共6页Application Research of Computers

基　　金：国家自然科学基金资助项目(71840003);上海市科委“科技创新行动计划”软科学重点项目(20692104300);上海理工大学科技发展资助项目(2018KJFZ043)。

摘　　要：传统随机森林填补方法并未考虑高维不平衡问题导致填补没有针对性,且使用0值预填补的方式可能会引入噪声并导致预测精度降低,因此提出一种基于Q学习和随机森林的缺失值填补方法(QL-RF)。该方法在特征选择后使用Q-learning权衡填补精度和填补数量,通过计算奖励筛选出具有填补价值的样本和特征组合,然后利用冗余特征填补重要特征中的缺失,并重点填补了少数类样本。此外,为提高不平衡数据下的分类效果,基于Bagging框架提出一种融合量子粒子群算法(QPSO)和XGBoost的集成分类模型(QXB)。实验表明:QL-RF在G-means、F_(1)-measure、AUC指标下均优于传统RF填补法,QXB显著优于SMOTE-RF和SMOTE-XGBoost,所提方法能够有效地处理高维不平衡数据下的缺失和分类问题。Traditional random forest filling method does not consider the problem of high-dimensional imbalance,which leads to filling untargeted.In addition,it need to prefill missing with 0,which may introduce noise and lead to the decrease in prediction accuracy.Therefore,this paper proposed a missing value filling method based on Q-learning and random forest(QL-RF).After feature selection,this method used Q-learning to weigh the filling accuracy and the filling quantity,then it selected the valuable samples and feature combinations by calculating reward,and it used the redundant features to fill the missing of important features with the minority samples filled mainly.Moreover,in order to improve the classification effect of imbalanced data,it proposed an ensemble classification model(QXB)based on bagging framework,which integrated quantum particle swarm optimization(QPSO)and XGBoost.The experiment results show that QL-RF is superior to traditional RF filling method in terms of G-means,F_(1)-measure and AUC,and QXB is significantly superior to SMOTE-RF and SMOTE-XGBoost.The proposed methods can effectively deal with the missing and classification problems under high-dimensional imbalance data.

关 键 词：高维不平衡 QL-RF BAGGING QPSO XGBoost QXB 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]