机构地区: 暨南大学信息科学技术学院数学系
出 处: 《南方经济》 2006年第3期113-119,共7页
摘 要: 本文利用线性判别分析建立信用评价模型,用来对中国2000年106家上市公司2000年96家上市公司进行两类模式(“好”、“差”)分类及三类模式(“好”、“中”、“差”)分类。对于线性判别分析法,又使用两种不同的方法进行判别分析:一种是利用SPSS统计软件对数据样本进行判别分析,称为LDA-SPSS方法;一种是利用原始样本数据推导建立线性判别分析模型,然后根据模型计算得到的结果对数据样本进行判别分析,称为LDA方法。仿真结果表明,无论是两类模式分类还是三类模式分类,LDA-SPSS的判别效果均优于LDA。但与多层感知器(MLP)相比,对两类模式分类,LDA-SPSS(100%)优于MLP(98.11%),MLP又优于LDALDA(95.28%);对三类模式分类,LDA-SPSS(91.67%)优于LDA(82.29%),LDA又优于MLP(79.17%)。 The paper uses linear discriminant analysis to establish two credit risk evaluation models. The two models are used to classify the 106 Chinese listed companies into two patterns: "good" and bad" and classify 96 Chinese listed companies into three patterns: "good", "median" and "bad". For the linear discriminant analysis, we also use two different methods. One is called LDA. The other is called LDA-SPSS. The simulating results show that the diseriminant effects of LDA-SPSS is better than that of LDA no matter the two patterns classification or the three patterns classification are used. However, as compared with miltilayer perceptron (MLP), for the two patterns classification, LDA-SPSS (100%) is better than MLP (98.11%) while MLP is better than LDA (95.28%). For the three patterns classification, LDA-SPSS (91.67%) is better than LDA (82.29%) while LDA is better than MLP (79.17%).