|
Abstract:
|
Classification of micro-array is an important aspect of cancer diagnosis and treatment. The support vector machine (SVM) and penalized logistic regression (PLR) have been successfully applied to micro-array problems. In this paper, we propose nu-ridge regression (nu-RR) as an alternative method to the SVM and PLR for the micro-array classification. Due to the goal in micro-array cancer diagnosis is mostly to identify the most responsible genes, rather than classification alone, we propose a new ranking method MR besides applying UR and RFE for gene selection. Our simulation on the real dataset, Leukemia diagnosis data, indicates nu-RR combined with either UR, MR or RFE tends to select less significant genes than other methods. Meanwhile, nu-RR performs superior to SVM and PLR with a lower rate in both cv-error and test error.
|
