Abstract: In classification problem, there are a large number of redundant and irrelevant features in datasets which  not only can??t increase the accuracy of classification, but also reduce the speed of the classification algorithm. Feature selection can solve these problems by maximizing the classification accuracy and minimizing the number of features, and it can be regarded as a multi-objective optimization problem because these are two contradictory objects. In order to enhance the efficiency of feature selection, an improved multi-objective artificial bee colony algorithm based on Knee Points for feature selection is proposed in this paper. A method for fast recognizing Knee Points is designed to improve the employed and onlooker bee phase. Experiment results on feature selection on 11 UCI datasets with 3 other traditional multi-objective algorithms show that the algorithm proposed in this paper has significant effect in reducing the number of classification and increasing the accuracy of classification.

Key words: multi-objective artificial bee colony algorithm, feature selection, Knee Points, classification algorithm

摘要： 在分类问题中，数据之间存在的大量冗余、不相关的特征不仅不能增加分类准确率，反而会降低分类算法执行的速度。特征选择通过最大化分类正确率和最小化特征数来解决这个问题，由于这是两个相互矛盾的目标，因此可以将特征选择问题视为一种多目标优化问题。为了提升特征选择的效率，提出了一种基于Knee Points的改进多目标人工蜂群算法的特征选择方法（artificial bee colony algorithm based on Knee Points， KnABC），设计了一种快速识别Knee Points的方法，并改进了引领蜂和引领蜂算子。与其他经典多目标算法在11个UCI测试数据集上的特征选择实验结果表明，提出的算法在减小分类特征数、增大分类结果准确率方面具有显著效果。

关键词: 多目标人工蜂群算法, 特征选择, Knee Points, 分类算法