In order to overcome the problem of high-dimensionality of gene expression data, a linear-based method for dimensionality reduction was proposed. Using Locality Preserving Projections (LPP) incorporated with the class information, the gene expression data were mapped into a feature subspace. Different from Principal Component Analysis (PCA) which searches the direction of maximal variance, CPP seeks a feature subspace that reflects the class information of the samples and by using CPP the dimensions of the data can be reduced while preserving the class information. Experiment results on real gene expression data compared with the classical linear technology PCA showed that CPP can identify the class feature better and obtain much better efficiency of visualization after dimensionality reduction via CPP than PCA.
