We found a match
Your institution may have rights to this item. Sign in to continue.
- Title
Relevant Component Locally Linear Embedding Dimensionality Reduction for Gene Expression Data Analysis.
- Authors
Xiaoping Min; Hai Wang; Zhiwei Yang; Shengxiang Ge; Jun Zhang; Ningxia Shao
- Abstract
Gene expression data typically contain many genes, which generates a feature vector with high dimensionality and considerable irrelevant information. Moreover, datasets typically contain few samples, which leads to the 'curse of dimensionality'. This dimensionality degrades classification performance. Therefore, it is essential to determine a good method for reducing the feature set size. In this paper, we proposed a relevant component locally linear embedding (RLLE) algorithm. This method changes the feature space used for data representation through a global linear transformation that assigns large weights to relevant dimensions and low weights to irrelevant dimensions. Next, the new distance between the sample points is calculated. Through processing the feature space, sample points of the same class could be efficiently neighbored. Based on the new distance, a locally linear embedding algorithm was applied to reduce the dimensionality of the samples. We applied the techniques to six published DNA microarray data sets and compare the RLLE algorithm with PCA, ISOMAP, LLE and RELIEFF using Naive Bayes classifiers. The experimental results and comparisons demonstrate that the proposed method is highly effective.
- Subjects
GENE expression; DATA analysis; DIMENSION reduction (Statistics)
- Publication
Metallurgical & Mining Industry, 2015, Issue 4, p186
- ISSN
2076-0507
- Publication type
Article