摘要
在高维数据空间中,存在大量冗余或无用的属性,这使得在子空间中寻找目标类更为有效.为此文章提出基于类别基尼系数子空间的加权互k近邻算法,利用类别基尼系数求出其对应的软子空间并将待分类样本和训练样本投影到各个类别子空间中,再在各软子空间中使用类别基尼系数加权距离互k近邻算法计算出待分类样本在各个子空间的投票权重并叠加,最终得出待分类样本的类标签.在公共数据集上的实验结果验证了该方法的有效性.
In high-dimensional data spaces, there exists a large number of redundant or useless attributes, and therefore it might be more effective to find target class in their subspaces. A weighted MKnn algorithm based on the Category＇s Gini Coefficient subspace is proposed in this paper. Using the Category＇s Gini Coefficient, the algorithm firstly calculates the corresponding soft subspaces, and projects the training and testing samples onto each category subspaces. Secondly, it calculates the vote weights of unclassified samples on each subspace by the weighted MKnn algorithm and then accumulates them. Finally, it obtains the category labels of unclassified samples. The experimental results on some UCI public datasets demonstrate the effectiveness of the proposed method.
出处
《计算机系统应用》
2014年第2期137-141,132,共6页
Computer Systems & Applications
基金
国家自然科学基金(61070062)
福建高校产学合作科技重大项目(2010H6007)
福建省教育厅B类项目(JB12201)
