Ternary Decision Trees with Locally-Adaptive Uncertainty Zones
Researchers have introduced ternary decision trees, which enhance standard binary decision trees by incorporating an uncertainty zone around decision boundaries. This zone allows for weighted blending of predictions from child subtrees and flags uncertain instances for different downstream handling. Five novel methods for estimating the uncertainty zone's width were proposed and evaluated, demonstrating significant improvements in accuracy over traditional CART methods across numerous datasets. AI
IMPACT Introduces a novel method for decision trees that improves accuracy by explicitly modeling uncertainty at decision boundaries.