摘要: 从概率密度函数的角度出发,利用Parzen窗法估计总体样本的概率密度分布,将核方法和Parzen窗法引入最大后验概率方法中,提出一种基于Parzen核估计的最大后验概率的高性能多分类方法。该方法不需要考虑样本数据的具体分布情况,能够得到分类的可信度,给出推理的不确定性依据。在3个国际标准UCI数据集和3个人脸数据集上的实验结果表明,该方法具有较好的分类效果。
关键词:
核函数,
arzen窗,
正态分布,
最大后验概率,
贝叶斯分类,
可信度
Abstract: From the perspective of probability density function, Parzen window estimation is adopted to estimate the probability density distribution of the overall samples. A novel Maximum A Posteriori(MAP) classification method based on Parzen kernel estimation which shows higher performance in the multi-way classification is proposed by introducing kernel method and the Parzen window method. It is not necessary to consider the specific distribution of the sample data in this proposed method, but the reliability for the classification is obtained and the evidence for uncertain reasoning is also available. Experimental results of the three international standard UCI data sets and three facial image data sets show that the method has better classification effect.
Key words:
kernel function,
Parzen window,
normal distribution,
Maximum A Posteriori(MAP),
Bayesian classification,
reliability
中图分类号:
张如艳, 王士同, 高恩芝. 基于Parzen核估计的最大后验概率分类方法[J]. 计算机工程, 2011, 37(16): 202-205.
ZHANG Ru-Yan, WANG Shi-Tong, GAO En-Zhi. Maximum A Posteriori Classification Method Based on Parzen Kernel Estimation[J]. Computer Engineering, 2011, 37(16): 202-205.