Classifying microarray cancer datasets using nearest subspace classification
In this paper we implement and test the recently described nearest subspace classifier on a range of microarray cancer datasets. Its classification accuracy is tested against nearest neighbor and nearest centroid algorithms, and is shown to give a significant improvement. This classification system uses class-dependent PCA to construct a subspace for each class. Test vectors are assigned the class label of the nearest subspace, which is defined as the minimum reconstruction error across all subspaces. Furthermore, we demonstrate this distance measure is equivalent to the null-space component of the vector being analyzed.
Supplementary Proceedings [of the] Third IAPR International Conference on Pattern Recognition in Bioinformatics (PRIB 2008)
Pattern Recognition and Data Mining