Classifying the decision to perform surgery in MEN1 cancer patients using decision trees

We present a user-friendly decision tree generating algorithm which searches for the best tree from a forest of n trees. We have biased our algorithm to create bigger trees, as they provide insight in the dataset's underlying structure. The algorithm was applied to data from 130 patients, who suffer from an hereditary form of cancer: MEN1. The best of multiple trees was picked based on performance that was evaluated by the algorithm, not only based on classification results of the validation test set, but also on the stability of the train- and validation test set accuracies, discriminative power, tree-width and tree-depth. We present a tree with performance 0.912 on a 0-1 scale, that closely resembles the decision to perform surgery in MEN1 patients by the physician. We also show that even good trees can make medically flawed decisions; that is why they must always be evaluated by health care professionals.