A counterfactual and risk temporal knowledge graph framework for interpretable project risk management

Effective project risk management (PRM) necessitates accurate prediction and actionable insights. Machine learning (ML) models improve risk assessment by uncovering complex patterns; however, their black-box nature limits interpretability, making it difficult for stakeholders to trust predictions. Traditional explainable artificial intelligence (XAI) methods highlight influential risk factors yet often fail to provide actionable recommendations, focusing on model behavior rather than practical interventions. Counterfactual explanations (CEs) aim to bridge this gap by suggesting modifications to risk factors or project conditions that could alter outcomes. However, existing CE methods in PRM often lack domain specificity, overlook interdependencies, and ignore temporal constraints, producing recommendations that are unrealistic or infeasible. To address these limitations, we propose Counterfactual Reasoning with Risk Temporal Knowledge Graph (CR-RTKG), a framework that integrates counterfactual reasoning with a Risk Temporal Knowledge Graph (RTKG) to improve interpretability and actionability of risk mitigation. The RTKG encodes domain knowledge, models causal dependencies and cascading effects, and classifies risks by temporal horizon, supporting prioritization based on urgency and systemic influence. By embedding stakeholder-defined constraints into a multi-objective optimization process, CR-RTKG generates context-sensitive and feasible counterfactuals. Unlike conventional methods, it aligns recommendations with real-world project constraints. Experimental results show that CR-RTKG achieves higher plausibility (96%) and feasibility (93%), outperforming baselines including Diverse Counterfactual Explanations (DiCE) and Flow-based Counterfactual Explanation (CeFlow).