用于操作风险分析小样本贝叶斯网络结构学习.pdfVIP

第 17 卷第4 期 Vol. 17 N o. 4 2008 年8 月 Journ al of Syst ems Man agement Aug. 2008 : 100522542( 2 008) 0420448207 1a b 2 1c ( 1. 上海立信会计学院a. 信息科学系 b . 中国立信风险管理研究院 c. 金融学系 上海201620; 2. 青岛大学经济学院 青岛2 6607 1 ) = 、 、d2sep aration : ; ; ; : T P 181 : A Learn in g Bayesian Networks St ruct ure from Small Dat a Set in Oper at ional Risk Analysis WAN G Shua ng2cheng 1a b , LI U Xi2hua 2 , ZH AN G P i2qiang 1c ( a. Depart ment of In for mat ion Scien ce b. R isk M anagement Research In stit ute c. Depart ment of F inan ce 1. Sh angh ai Lix in Un iver sit y of Comm er ce Sh anghai 2 01600 Ch in a; 2 . Econ omic Inst it ut e Qin gdao Un iver sit y Qin gdao 266071 Ch in a) =Ab strac t At present th e meth od s of lear nin g Bay esian net work s st ruct ure n eed a large number of dat a wit h h igh qualit y. T he algorit hm s have low efficiency an d reliab ility . But it is v ery difficu lt t o accumu lat e many reliable ex amp les in op eration al risk man agem ent . In t h is p aper a n ew m et h od of learn in g Bayesian net work s st ru ct ure from sm al l dat a set is present ed based on b asic depen dency r elat ion sh ip bet ween varia2 b les b asic st ru cture b etween n odes d2sep aration crit erion an d depen dency analysis met hod. It can effec2 t ively avoid t h e problems ab ove. Th e exp er im ent s an d analysis are don e by using stimu lant an d real2world dat aset. Experiment al resu lts show t hat t h is met h od can effect ively learn Bay