基于多元密度核估计的Bayes判别法  

Bayes Multivariate Discriminant Method Based on Kernel Density Estimation

作　　者：陈相兵[1] 

机构地区：[1]四川大学锦江学院,四川彭山620860

出　　处：《数学的实践与认识》2015年第10期185-189,共5页Mathematics in Practice and Theory

摘　　要：判别分析是根据所研究个体的观测指标来推断该个体所属类型的一种统计方法,但一般的Bayes判别法都须假设各个判别变量服从正太分布,且经常出现回代误判率不为零的问题.首次将多元密度核估计应用到Bayes判别法中,消除了判别变量要服从正太分布这个条件约束,并应用建立的最优窗宽调整算法,将回代误判率调整为零,最后对实例进行分析看出结论应用效果较好.Discriminant analysis is a statistical method based on the observation index of individual to infer the individual type.Generally, the Bayes discriminant method must assume all discriminant variables obey normal distribution and back substitution error rate always be not zero. In this paper, the multivariate density kernel estimation is applied to the Bayes Discriminant method firstly, and eliminates the discriminant variables with this constraint for obeying normal distribution. Appling optimal bandwidth adjustment algorithm in this paper, adjust the back substitution error rate to zero. Finally, the conclusion of this paper shows good application effect.

关 键 词：平均误判损失 密度核估计 Bayes判别法 最优窗宽 

分 类 号：O212.8[理学—概率论与数理统计]