Temporary repository for our paper: "Counterfactual Explanation Trees: Transparent and Consistent Actionable Recourse with Decision Trees," to apper in AISTATS-22.
CET is a framework of Counterfactual Explanation (CE) that summarizes actions over the entire input space by a decision tree.
Basic running examples are provided in demo.py:
$ python demo.py
# CET Demonstration
* Dataset: EmployeeAttrition
* x_1 : Age (I:FIX)
* x_2 : BusinessTravel (I)
* x_3 : Education (I:FIX)
* x_4 : JobLevel (I:FIX)
* x_5 : MonthlyIncome (I)
* x_6 : OverTime (B)
* x_7 : PercentSalaryHike (I)
* x_8 : OutstandingPerformanceRating (B:FIX)
* x_9 : TotalWorkingYears (I:FIX)
* x_10: TrainingTimesLastYear (I:FIX)
* x_11: WorkLifeBalance (I)
* x_12: YearsAtCompany (I:FIX)
* x_13: YearsInCurrentRole (I:FIX)
* x_14: YearsSinceLastPromotion (I:FIX)
* x_15: YearsWithCurrManager (I:FIX)
* x_16: Department:HumanResources (B)
* x_17: Department:ResearchAndDevelopment (B)
* x_18: Department:Sales (B)
* x_19: JobRole:HealthcareRepresentative (B)
* x_20: JobRole:HumanResources (B)
* x_21: JobRole:LaboratoryTechnician (B)
* x_22: JobRole:Manager (B)
* x_23: JobRole:ManufacturingDirector (B)
* x_24: JobRole:ResearchDirector (B)
* x_25: JobRole:ResearchScientist (B)
* x_26: JobRole:SalesExecutive (B)
* x_27: JobRole:SalesRepresentative (B)
* Classifier: LightGBM
* n_estimators: 100
* num_leaves: 16
* train score: 0.9809437386569873
* train denied: 160
* test score: 0.845108695652174
* test denied: 33
## Counterfactual Explanation Tree (CET)
* Parameters:
* lambda: 0.01
* gamma: 1.0
* max_iteration: 100
* leaf size bound: 101
* LIME approximation: True
* leaf size: 3
* Time[s]: 23.906064433
### Learned CET
- If OverTime:
* Action [Attrition: Yes -> No] (19/22 = 86.4% / MeanCost = 0.274):
* OverTime: True -> False
- Else:
- If BusinessTravel<2:
* Action [Attrition: Yes -> No] (5/6 = 83.3% / MeanCost = 0.173):
* MonthlyIncome: +1282
- Else:
* Action [Attrition: Yes -> No] (4/5 = 80.0% / MeanCost = 0.187):
* BusinessTravel: -1
@InProceedings{Kanamori:AISTATS2022,
title = {{Counterfactual Explanation Trees: Transparent and Consistent Actionable Recourse with Decision Trees}},
author = {Kanamori, Kentaro and Takagi, Takuya and Kobayashi, Ken and Ike, Yuichi},
booktitle = {Proceedings of the 25th International Conference on Artificial Intelligence and Statistics},
pages = {1846-1870},
year = {2022},
}