文摘
An important application of gene expression data is tumor classification. Dimensionality reduction is a key step of tumor classification as gene expression data have the so-called large and small problem. To reduce the dimensionality of the microarray data, in this paper, a novel algorithm called orthogonal discriminant neighborhood analysis (ODNA) is proposed for tumor classification, which can reduce the effect resulting from over-fitting by pre-selecting a small subset of genes. Given a set of data points in the ambient space, a neighbor weight matrix is firstly built to describe the relationship among the data samples. Secondly, optimal between-class scatter matrix and within-class scatter matrix are defined such that the neighborhood structure can be preserved. To improve the discriminating ability, a new method is presented to orthogonalize the basis eigenvectors. The experimental results with two public microarray datasets demonstrate that the proposed ODNA is quite effective and feasible for tumor classification. Keywords Locality sensitive discriminant analysis (LSDA) Microarray data Orthogonal discriminant neighborhood analysis (ODNA) Tumor classification