Skip to main content

Table 4 Comparison of various classifiers in cancer prognosis datasets

From: Top scoring pairs for feature selection in machine learning and applications to cancer outcome prediction

Dataset Error rate on 2X 5-fold CV (%) Error rate on the test set (%)
  TSP k- TSP SVM k -TSP+SVM TSP k -TSP SVM k -TSP+SVM
van't Veer Breast cancer      68.4 47.3 31.6 10.5
Wang Breast cancer 41.4 ± 2.5 37.3 ± 2.8 30.1 ± 1.8 32.9 ± 3.0     
Lung adenocarcinoma 41.2 ± 2.5 40.7 ± 2.5 29.7 ± 3.5 27.8 ± 2.9     
Medulloblastoma 48.3 ± 4.3 49.6 ± 7.4 37.5 ± 5.9 35.8 ± 4.8     
  1. In the van't Veer breast cancer dataset where there is an independent test set, the error rate on the test set was obtained at the gene selection level at which the training set achieves its minimum LOOCV error rate. In the other datasets where there is no separate test set, the error rates (mean ± SE) were obtained from two experiments of five-fold cross validation.