Supplementary information:

New Feature Subset Selection Procedures for Classification of Expression Profiles

Trond Hellem Bø and Inge Jonassen

This page contains supplementary information to the paper New Feature Subset Selection Procedures for Classification of Expression Profiles. Here you will find all the figures not included in the paper regarding prediction accuracies. Also included are links to web-adresses from which the datasets we have used can be downloaded.

Results/Plots

The plots below are of average prediction accuracy performance, using three prediction methods and holding back different portions of the data from the training set. LOOCV stands for Leave-One-Out Cross Validation, L-24-OCV for Leave-24-Out Cross Validation, and so on. Each curve show the performance for feature sets of varying size selected by four feature selection methods.

ALL/AML dataset

The ALL/AML dataset was publised by Golub et al. (Science, 1999). Our results were achieved using all 72 experiments.

Plot using 3NN prediction and LOOCV	Plot using 3NN prediction and L-24-OCV	Plot using 3NN prediction and L-36-OCV
Plot using DLD prediction and LOOCV	Plot using DLD prediction and L-24-OCV	Plot using DLD prediction and L-36-OCV
Plot using FLD prediction and LOOCV	Plot using FLD prediction and L-24-OCV	Plot using FLD prediction and L-36-OCV

Colon dataset

The Colon dataset was published by Alon et al. (PNAS, 1999).

Plot using 3NN prediction and LOOCV	Plot using 3NN prediction and L-20-OCV	Plot using 3NN prediction and L-31-OCV
Plot using DLD prediction and LOOCV	Plot using DLD prediction and L-20-OCV	Plot using DLD prediction and L-31-OCV
Plot using FLD prediction and LOOCV	Plot using FLD prediction and L-20-OCV	Plot using FLD prediction and L-31-OCV