FOCUS: Flexible Optimizable Counterfactual Explanations for Tree Ensembles
نویسندگان
چکیده
Model interpretability has become an important problem in machine learning (ML) due to the increased effect algorithmic decisions have on humans. Counterfactual explanations can help users understand not only why ML models make certain decisions, but also how these be changed. We frame of finding counterfactual as optimization task and extend previous work that could applied differentiable models. In order accommodate non-differentiable such tree ensembles, we use probabilistic model approximations framework. introduce approximation technique is effective for predictions original show our examples are significantly closer instances than those produced by other methods specifically designed ensembles.
منابع مشابه
Explanations of Counterfactual Inferences
When engaging in counterfactual thought, people must imagine changes to the actual state of the world. In this study, we investigated how people reason about counterfactual scenarios by asking participants to make counterfactual inferences about a series of causal devices (i.e., answer questions such as If component X had not operated [had failed], would components Y, Z, and W have operated?) a...
متن کاملCausal Explanations in Counterfactual Reasoning
This paper explores the role of causal explanations in evaluating counterfactual conditionals. In reasoning about what would have been the case if A had been true, the localist injunction to hold constant all the variables that causally influence whether A is true or not, is sometimes unreasonably constraining. We hypothesize that speakers may resolve this tension by including in their delibera...
متن کاملusing counterfactual analysis for providing historical explanations in social sciences
counterfactual analysis is concerned with explaining events that have not happened. counterfactuals are mental experiments through which one can reconstruct hypothetical versions of the history in one’s mind; these versions are relatively different from the real history, but provide one with the opportunity to test historical hypotheses against the available evidence. historicist researchers in...
متن کاملOptimization of Tree Ensembles
Tree ensemble models such as random forests and boosted trees are among the most widely used and practically successful predictive models in applied machine learning and business analytics. Although such models have been used to make predictions based on exogenous, uncontrollable independent variables, they are increasingly being used to make predictions where the independent variables are cont...
متن کاملConsistent Individualized Feature Attribution for Tree Ensembles
Interpreting predictions from tree ensemble methods such as gradient boosting machines and random forests is important, yet feature attribution for trees is often heuristic and not individualized for each prediction. Here we show that popular feature attribution methods are inconsistent, meaning they can lower a feature’s assigned importance when the true impact of that feature actually increas...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: Proceedings of the ... AAAI Conference on Artificial Intelligence
سال: 2022
ISSN: ['2159-5399', '2374-3468']
DOI: https://doi.org/10.1609/aaai.v36i5.20468