Skip to main content
. 2011 Apr 7;6(4):e18007. doi: 10.1371/journal.pone.0018007

Figure 3. Classification accuracy.

Figure 3

Classification accuracy of our complete leave-one-out cross-validation experiment at all nodes of our decision tree. Five different panel sizes are evaluated, with 30K corresponding to all available markers, 2K corresponding to the top 2,000 PCAIMs, and P1, P2, and P3 corresponding to the panel sizes depicted in Table 1. These smaller panels emerged by removing redundant markers from the top 2,000 AIMs. Notice that the top 2,000 markers were selected using only the individuals in the training set of the crossvalidation experiment. (A) Classification accuracy results (out of 100%). (B) Average number of correctly predicted nearest neighbors (out of five).