Abstract:

Precise cancer classification is essential to the successful diagnosis and treatment of cancers.Al-though semi- supervised dimensionality reduction approaches perform very well on clean data sets,the topology of the neighborhood constructed with most existing approaches is unstable in the presence of noise.In order to solve this problem,a novel random subspace- based semi- supervised dimensionality reduction algorithm marked as RSSSDR,which combines the random subspace with the semi- supervised dimensionality reduction,is pro-posed.In this algorithm,first,multiple diverse graphs are designed in different random subspaces of data sets and are then combined to form a mixture graph on which dimensionality reduction is performed.Subsequently,the edge weights of neighborhood graph are determined through minimizing the local reconstruction error,such that the global geometric structure of data can be preserved without changing the local geometric structure.Ex-perimental results on public cancer data sets demonstrate that the proposed RSSSDR algorithm is of high classifi-cation accuracy and strong robustness.

Key words: semi- supervised learning, random subspace, cancer classification, dimensionality reduction