数据流分类中的增量特征选择算法

Incremental feature selection algorithm for data stream classification

概念流动的出现及数据的高维性增加了数据流特征选择的复杂性。信息增益是最有效的特征选择算法之一,但计算量大。对信息增益做了等价替换,提出一种基于改进信息增益的混合增量特征选择(IFS)算法。该算法首先利用与分类器无关的评价函数选出候选特征集合,然后将分类器作用于候选特征集合,利用分类精度作为评价标准去选择特征子集,在遇到概念漂移时重新选择特征子集。通过在超平面数据集和UCI数据集上的实验,表明基于IFS算法的分类器能够很快地适应概念漂移,并且比基于全部特征的分类算法有更高的精度。

The complexity of feature selection for real-world data stream will increase because of high-dimensional data and concept drifting. Information gain is one of the most effective feature selections, but its computation is too huge. In order to deal with the problem, the authors proposed an incremental feature selection algorithm based on improved information gain, named IFS. Firstly, the algorithm selected candidate feature set by using independent evaluation function; secondly, feature set was selected with classifer role in candidate feature set. Finally, it selected feature set again while encountering concept drifting. The experiment was operated on moving hyperplane data set and UCI data set. The experimental results show that the proposed approach can adapt to the concept drifting with higher speed and works much better than non-feature selection algorithms.