基于粒子群优化聚类算法的多预测器信用评估模型  

A Multi-predictor Credit Evaluation Model Based on Particle Swarm Optimizing Clustering Algorithm

作　　者：张润驰 杜亚斌[1] ZHANG Run-chi,DU Ya-bin(Department of Finance,School of Business, Nanjing University, Nanjing 210093 ,Chin)

机构地区：[1]南京大学商学院金融系,江苏南京210093

出　　处：《系统工程》2017年第10期154-158,共5页Systems Engineering

基　　金：国家自然科学基金重大研究计划项目(90718008);温州大学金融研究院招标金改研究项目(RH1206058)

摘　　要：经典的加权k均值聚类算法能够有效区分不同属性对聚类过程的影响程度,但同时也易因权值的选取不当导致预测性能较差。本文在其基础上,针对信用评估问题,设计了多预测器粒子群优化加权k均值聚类(MPWKM)模型。MPWKM模型首先对样本数据进行预处理,剔除重要程度较低的属性,接着以粒子群算法搜索加权k均值聚类算法的最优权值组合,解决权值选择问题,进而构建多个基于不同样本空间子集的基预测器,最后根据各基预测器的预测结果组合成完整的预测模型,进一步提升模型的性能。实证研究表明:MPWKM模型与现有的五个成熟模型相比,在预测精度较高的同时,也具有较好的平衡性与稳定性。The classical weighted k-means clustering algorithm can effectively distinguish the impact of different attributes in the clustering process,but it tends to get a poor classification performance due to the improper selection of weights of each attribute.According,we design a multi-predictor particle swarm optimizing w-k-means clustering（MPWKM）model to solve credit evaluation problem.By using MPWKM model we firstly perform the pretreatment on the sample data to eliminate the less important attributes,then search the optimal weight combination of the weighted k-mean clustering algorithm by particle swarm optimization algorithm to solve the weight selection problem,and further construct several basic predictors based on the different subset of the original sample space.We finally construct the complete model according to the prediction accuracy of each basic predictor to further improve the performance of the model.Empirical study shows that the MPWKM model has high prediction accuracy,and better balance and stability compared with the existing five mature credit evaluation models.

关 键 词：信用评估 MPWKM模型 粒子群算法 加权k均值聚类算法 

分 类 号：F830[经济管理—金融学]