Decision Tree Learning for Uncertain Clinical Measurements

Clinical decision requires reasoning in the presence of imperfect data. DTs are a well-known support tool, owing to their interpretability, fundamental safety-critical contexts such as medical diagnosis. However, learning from uncertain data leads poor generalization, and generating predictions for hinders prediction accuracy. Several methods have suggested potential probabilistic decisions at internal nodes making robust uncertainty. Some approaches only employ thresholds during evaluation. Others also consider uncertainty phase, expense increased computational complexity or reduced interpretability. The existing not clarified merit approach distinct phases DT learning, nor when is present training test We that models measurement noise distribution, independently realized: (1) searching split thresholds, (2) splitting instances, (3) unseen soft (1, 2) achieved regularizing effect, leading significant reductions size, while maintaining accuracy, noise. Soft evaluation showed no benefit handling