基于邻域覆盖约简的层次化规则学习算法

Hierarchical Rule Learning Algorithm Based on Neighborhood Covering Reduction

在大数据环境下的分类学习中,随着描述样本语义信息的丰富,数据的类别空间结构存在着层次化。然而,现有分层分类算法缺乏可解释性,为此本文提出了一种基于邻域覆盖约简的层次化规则学习算法。该算法框架主要包括:(1)定义了面向层次化结构数据的邻域覆盖约简模型;(2)定义了层次邻域系统中覆盖元的依赖度;(3)提出了一种基于覆盖元依赖度的层次化规则学习前向搜索算法。最后,实验表明本文所提算法的分类性能较优且具有较好的可解释性。

In the classification learning under big data environment,there are hierarchical structures among categories in class space of data with rich semantic information.However,existing hierarchical classification algorithms lack interpretability.To address this problem,a hierarchical rule learning method based on neighborhood covering reduction is proposed.The framework consists of three parts:(1)a neighborhood covering reduction model for hierarchical structure data is defined;(2)a dependency degree of covering element in hierarchical neighborhood system is defined;(3)a forward hierarchical rule learning algorithm based on the dependency degree of covering element is designed.Finally,the effective and better interpretability of the proposed algorithm is verified by extensive experiments.