Figure 5From: Top scoring pairs for feature selection in machine learning and applications to cancer outcome predictionComparison of TSP, Fisher and RFE as feature selection methods in the cancer prognostic datasets. A) shows the SVM and KNN classification error rates on the test set of van't Veer Breast cancer dataset at various gene selection levels, using TSP, Fisher and RFE as feature selection methods. B) shows LOOCV error rates by SVM in Lung adenocarcinoma and Medulloblastoma datasets at various gene selection levels, using TSP, Fisher and RFE as feature selection methods. The x-axis is the number of top ranked gene pairs for TSP, or half the number of top ranked genes for Fisher and RFE. The horizontal lines are the error rates of SVM or KNN using all features.Back to article page