Figure 2

The genetic programming process. This iterative technique was employed on the training set samples to generate classifier rules that were tested on the validation set. Randomly chosen components were initially used to create a population of candidate programs from which a small mating pool of candidate programs was generated. Inputs were passed into these programs and the predicted nodal statuses were evaluated for fitness. The two best performing programs were then mated to produce offspring that replaced the two least fit programs. This process was repeated over many generations to create better programs.