基于模糊粗糙集的层次分类增量特征选择

随着大数据时代的到来,数据的类标签数量急剧增加,对现有的分类任务带来了严峻的挑战。为了解决这个问题,人们通常将标签组织成层次结构,使用结构中所包含的信息来对任务进行学习。考虑样本的不断增加,使用模糊粗糙集信息熵设计了一种面向层次分类的增量特征选择算法。考虑兄弟策略,将现有的λ条件熵推广到了层次分类的情形,设计了一种非增量的层次分类特征选择算法,设计了λ增量条件熵,基于此设计了增量版本的特征选择算法。在实验中,采用了包括非增量版本在内的7种不同的特征选择算法在5个层次数据集上与增量算法进行比较,实验结果验证了2种算法的有效性,并且所设计的增量算法能在不影响性能的情况下加快特征选择的进程。

Incremental feature selection for hierarchical classification based on fuzzy rough sets

The number of class labels has increased dramatically in the age of big data, posing a serious challenge to existing classification tasks. To solve this problem, people typically organize the labels into a hierarchical structure and then use the information in the structure to learn the task. Considering the increasing number of samples, an incremental feature selection algorithm for hierarchical classification is designed by using fuzzy rough sets-based information entropy. Firstly, considering the sibling strategy, the existing λ conditional entropy is generalized to the case of hierarchical classification, and a non-incremental hierarchical classification feature selection algorithm is designed. Secondly, the λ incremental conditional entropy is defined, based on which an incremental version of the feature selection algorithm is designed. In the experiment, seven different feature selection algorithms, including the non-incremental version, are used to compare with the incremental algorithm on five hierarchical datasets. The experimental results verify the effectiveness of the two algorithms, and the design incremental algorithm can accelerate the process of feature selection without affecting performance.