两种基于K近邻特征选择算法的对比分析

A comparison between two KNN based feature selection algorithms

在特征选择过程中,针对近邻错误分类率较低的问题,分别采用正向贪心和逆向贪心思想设计了两种启发式特征选择算法,其目的是在降低数据集中特征数量的同时,能够进一步降低近邻错误分类率。通过8组UCI数据集上的交叉验证结果表明,相比于正向贪心算法,逆向贪心算法能够删除较多的冗余特征,从而得出逆向贪心算法能够更有效地提高近邻算法的分类精度的结论。

To reduce the nearest neighbor error classification rate in feature selection, two heuristic algorithms are proposed by using the forward and reverse greedy ideas, respectively. The purpose of our paper is not only to reduce the features in data sets, but also to further reduce the nearest neighbor error classification rate. By the cross validation on eight UCI data sets, the experimental results show that the reverse greedy algorithm can not only remove more redundant features, but also can effectively improve the classification accuracy of the nearest neighbor algorithm.