Scholarlink

Explainable Artificial Intelligence (XAI) models are of growing interest as they provide explanations to enhance the AI models’ transparency. For instance, the Counterfactual Explanations (CE) model is an optimisation model used in case of unpleasant machine learning (ML) outcomes to recommend countering the results into desirable ones with minimal changes. The main challenge of CE models in the literature is finding feasible solutions. This paper proposes the Long-term Counterfactual Explanation (LTCE) to leverage infeasible suggestions within CE models, shedding light on hidden casualties. As a case study, we tested the risk management system of a maritime container terminal’s logistics, and infeasible solutions suggested hidden vulnerabilities and issues in the system. The experiments revealed that while 22% of the records have no feasible solutions, the LTCE can recognise impactful, unmodifiable features such as weather for designing long-term strategies to enhance the system’s resiliency in the future. Especially temperature (29.11%) and air pressure (20.66%) were found to be the most significant features, highlighting the potential of LTCE in addressing resiliency vulnerabilities.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Developing Long-Term Business Strategies by Leveraging Infeasible Recommendations of the Counterfactual Explanation Model

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Amir Hossein Ordibazar,
  • Omar K. Hussain,
  • Ripon Kumar Chakrabortty,
  • Elnaz Irannezhad,
  • Morteza Saberi

摘要

Explainable Artificial Intelligence (XAI) models are of growing interest as they provide explanations to enhance the AI models’ transparency. For instance, the Counterfactual Explanations (CE) model is an optimisation model used in case of unpleasant machine learning (ML) outcomes to recommend countering the results into desirable ones with minimal changes. The main challenge of CE models in the literature is finding feasible solutions. This paper proposes the Long-term Counterfactual Explanation (LTCE) to leverage infeasible suggestions within CE models, shedding light on hidden casualties. As a case study, we tested the risk management system of a maritime container terminal’s logistics, and infeasible solutions suggested hidden vulnerabilities and issues in the system. The experiments revealed that while 22% of the records have no feasible solutions, the LTCE can recognise impactful, unmodifiable features such as weather for designing long-term strategies to enhance the system’s resiliency in the future. Especially temperature (29.11%) and air pressure (20.66%) were found to be the most significant features, highlighting the potential of LTCE in addressing resiliency vulnerabilities.