一种改进的基于模糊似然函数的特征选择算法

An Improved Algorithm of Feature Selection Based on Fuzzy Likelihood Function

本文在Shang模糊似然函数的基础上,提出了基于特征空间的可信度函数和特征分量影响函数以改进Shang模糊似然函数的可靠性,采用改进的模糊似然函数评价任意特征子集的有效性,可更真实地描述模式在特征空间的分布,是一种有效的特征选择算法。论文将改进算法与原算法进行了测试比较,结果表明改进的特征选择算法具有更好的可靠性和实用性。

The paper presents an improved algorithm of feature selection based on Shang fuzzy likelihood function. Reliability function and influence function of features are constructed to improve the efficiency of Shang fuzzy likelihood function. The improved algorithm, which truly describes the distribution of patterns in feature space, can be used to estimate the effectiveness of any subset of features for pattern recognition. The algorithm presented in this paper is compared with the original one. The result reveals that the improved algorithm has a better performance in reliability and practicability for feature selection.