Figure 6. Projection pursuit (PP). The first principal component (PC1) is rotated to an optimal separation of two groups of data points, thereby defining the first PP. From a random starting plane, two candidate planes are generated. The projection index is evaluated, using the method of Posse, to find the optimal PP index. The plane containing the best PP index is then selected as a new starting plane and the PP axes are identified from projection onto the two best planes.