Figure 1

Iterative feature removal on influenza and lung cancer data. Iterative Feature Removal is shown using two data sets, influenza (top) and lung cancer (bottom). In each row, the left figure shows the accuracy at each iteration and the right figure shows the number of features selected per iteration. At each iteration, the model is trained without access to any of the genes selected in any of the previous iterations. For the influenza data set there are about 40 sets that are approximately equally predictive identifying approximately 1200 genes. For the lung cancer data there are about 30 sets, or some 900 genes that exhibit predictive properties. The red line represents the rolling average accuracy, illustrating the trend in the data. Figure best viewed in color.