Intelligent Computing Theories and Methodologies

Abstract

In this paper, a discriminant manifold learning method based on Locally Linear Embedding (LLE), which is named Locally Linear Representation Fisher Criterion (LLRFC), is proposed for the classification of tumor gene expressive data. In the proposed LLRFC, an inter-class graph and intra-class graph is constructed based on the class information of tumor gene expressive data, where the weights between nodes in both graph are optimized using locally linear representation trick. Moreover, a Fisher criterion is modeled to maximize the inter-class scatter and minimize the intra-class scatter simultaneously. Experiments on some benchmark tumor gene expressive data validate its efficiency.