基于核的最小距离分类法的参数选择方法

Parameter Selection Method Based on Kernel Nearest Neighbor Classification

在基于核函数的最小距离分类方法对数据集进行分类过程中,目标函数的核函数参数选择直接影响分类器的分类成功率。该文提出一种选择应用目标函数来选择适当参数的方法。实验结果表明,与单纯的基于核的最小距离分类法相比,选择最优核函数的参数可以提高分类器的成功率。

This paper proposes kernel nearest neighbor to classify the data sets. The parameter of the kernel function is the most difficulty question of this research topic, so this paper proposes a hybrid approach to use the target function to choose an adaptive parameter. Through testing the approach on a typical classification data sets, and the preliminary results demonstrate that, the target function can provide an adaptive parameter to optimize the kernel function for classification in various domains, especially compared with other kernel-based nearest neighbor classification methods.