An identifiable cost-aware causal decision-making framework using counterfactual reasoning.

Area of Science:

• Artificial Intelligence
• Causal Inference
• Systems Engineering

Background:

• Decision-making under abnormal conditions is crucial for system restoration.
• Existing frameworks often neglect action costs or adequate causal mechanisms.
• Reinforcement learning and root cause analysis have limitations in these scenarios.

Purpose of the Study:

• Propose a minimum-cost causal decision (MiCCD) framework using counterfactual reasoning.
• Address challenges in identifying counterfactual reasoning with mixed anomaly data.
• Optimize intervention strategies in continuous decision spaces while considering costs.

Main Methods:

• Developed a surrogate model based on causal graphs with abnormal pattern clustering labels.
• Approximated the structural causal model for identifiable counterfactual reasoning.
• Employed the Sequential Least Squares Programming (SLSQP) algorithm for cost-aware optimization.

Main Results:

• MiCCD demonstrated superior performance across F1-score, cost efficiency, and ranking quality (nDCG@k).
• Effectiveness validated on both synthetic and real-world datasets.
• Outperformed conventional decision-making methods in abnormal condition management.

Conclusions:

• The MiCCD framework effectively integrates causal inference and cost optimization for abnormal condition decision-making.
• MiCCD offers a robust and broadly applicable solution for complex system management.
• This approach enhances the efficiency and cost-effectiveness of restoring systems to normal states.