PLS and SVD based penalized logistic regression for cancer classification using microarray data

Accurate cancer prediction is important for treatment of cancers. The combination of two dimension reduction methods, partial least squares (PLS) and singular value decomposition (SVD), with the penalized logistic regression (PLR) has created powerful classifiers for cancer prediction using microarray data. Comparing with support vector machine (SVM) on seven publicly available cancer datasets, the new algorithms can achieve very good performance and run much faster. They also have the advantage that the probabilities of predictions can be directly given. PLS based PLR is also combined with recursive feature elimination (RFE) to select a 16-gene subset for acute leukemia cancer classification. The testing error on this subset of genes is empirically zero.