193947.fig.003
Figure 3: Support Vector Machine classification of AML versus Normal. The classifier was trained with 21 AML and 21 Normal Instances (left-most two regions). The classifier was then used to blindly predict class probabilities for the test set of 21 AML and 288 Normal instances (the right-most region). Ground-truth class assignments are indicated by color, red for AML and blue for Normal. The probability range 0-1 was divided into three equal regions. Instances falling into the lower third were classified as AML, in the upper third as Normal, and in the middle as Uncertain “?”.