Researchers have developed a new method for explaining the predictions of tree ensemble models, which are widely used for tabular data. This approach, called a local Decision Predicate Graph (local DPG), captures the specific rules executed for a given sample, providing a verifiable record of the computation. The method identifies a "critical predicate"—the rule crossing that most influences the model's decision—offering users a concrete threshold and the associated trees without relying on external explainers. Experiments across multiple datasets and ensemble types show that this critical predicate improves the separation of differently predicted neighbors and can recover the true class in a significant percentage of incorrect predictions. AI
IMPACT Enhances interpretability for widely used tree ensemble models, aiding in debugging and trust.
RANK_REASON The item is an academic paper detailing a new method for explaining machine learning models. [lever_c_demoted from research: ic=1 ai=1.0]
- alphaXiv
- arXiv
- CatalyzeX
- DagsHub
- Decision Predicate Graphs
- Gotit.pub
- Hugging Face
- IArxiv
- ScienceCast
- Sylvio Barbon Junior
- Tree Ensembles on the Induced Discrete Space
- TreeSHAP
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →