论文部分内容阅读
精确的癌症分类对于癌症的成功诊断和治疗是必不可少的.半监督维数约减算法在干净的数据集上表现得很好,然而当面临噪声时,当前的大部分算法所构造的邻域结构是拓扑不稳定的.为了克服这一问题,文中提出了一种基于随机子空间的半监督维数约减算法(RSSSDR),将随机子空间与半监督维数约减算法结合起来.在数据集的不同随机子空间上,该算法首先设计多个不同的子图,然后将这些子图联合起来构成一个混合图并在其上进行维数约减.该算法通过最小化局部重构误差来确定邻域图的边权值,在保持癌症数据集局部结构的同时能够保持其全局结构.在公共癌症数据集上的实验结果表明,RSSSDR算法具有较高的分类准确率和较好的参数鲁棒性.
Precise cancer classification is essential for the successful diagnosis and treatment of cancer.Some semi-supervised dimensionality reduction algorithms perform well on clean data sets, however, when faced with noise, most of the current algorithms construct neighbors In order to overcome this problem, a semisupervised dimension reduction algorithm (RSSSDR) based on stochastic subspace is proposed in this paper, which combines the stochastic subspace with the semicarred dimension reduction algorithm. In different random subspaces of data set, the algorithm designs several different subgraphs first, and then combines the subgraphs to form a hybrid graph and reduces the dimension on it. This algorithm minimizes the local reconstruction Error to determine the edge weight of the neighborhood graph, while maintaining the local structure of the cancer dataset while maintaining its global structure.Experimental results on the public cancer dataset show that the RSSSDR algorithm has a higher classification accuracy and better Parameter robustness.