A counterfactual and risk temporal knowledge graph framework for interpretable project risk management
Bodrunnessa Badhon, Mario Vanhoucke, Sreenatha G. Anavatti, Ripon K. Chakrabortty
Engineering Applications of Artificial Intelligence
Problems Identified (5)
Black-box risk prediction interpretability: Machine learning risk assessment models are difficult for stakeholders to trust because their black-box nature limits interpretability.
Non-actionable XAI explanations: Traditional XAI methods identify influential risk factors but often do not provide practical intervention recommendations.
Infeasible counterfactual recommendations: Existing counterfactual explanation methods in PRM can lack domain specificity, overlook interdependencies, and ignore temporal constraints, leading to unrealistic or infeasible recommendations.
Black-box risk prediction interpretability: Machine learning risk assessment models are difficult for stakeholders to trust because their black-box nature limits interpretability.
Non-actionable XAI explanations: Traditional XAI methods identify influential risk factors but often do not provide practical intervention recommendations.
Proposed Solutions (5)
CR-RTKG framework: The paper proposes CR-RTKG, which integrates counterfactual reasoning with a Risk Temporal Knowledge Graph to improve the interpretability and actionability of risk mitigation.
Risk Temporal Knowledge Graph modeling: The RTKG encodes domain knowledge, causal dependencies, cascading effects, and temporal risk horizons to support prioritization by urgency and systemic influence.
Constraint-aware counterfactual optimization: CR-RTKG embeds stakeholder-defined constraints into multi-objective optimization to produce context-sensitive and feasible counterfactual explanations.
CR-RTKG framework: The paper proposes CR-RTKG, which integrates counterfactual reasoning with a Risk Temporal Knowledge Graph to improve the interpretability and actionability of risk mitigation.
Risk Temporal Knowledge Graph modeling: The RTKG encodes domain knowledge, causal dependencies, cascading effects, and temporal risk horizons to support prioritization by urgency and systemic influence.
Results (3)
Improved plausibility and feasibility:
Outperforms counterfactual baselines:
Improved plausibility and feasibility:
Research Domain
Project risk management and explainable AI with temporal knowledge graphs