An integrated explainable AI framework with counterfactual optimization for decision support
- 1Department of Business Administration and Management, Washington University of Science and Technology, Alexandria, Virginia, United States of America
- 2Department of CSE, Bangladesh University of Business and Technology, Dhaka, Bangladesh
- 3Information Technology, St. Francis College, Brooklyn, New York, United States
Abstract
Artificial intelligence is increasingly used to support managerial decision-making in domains such as human resource management and customer relationship management; however, the lack of transparency in complex predictive models limits their practical adoption. This paper proposes a counterfactual-guided explainable artificial intelligence framework that integrates deep learning, feature attribution, and counterfactual reasoning to deliver accurate, interpretable, and actionable managerial insights. The proposed model is evaluated on an employee attrition dataset and a customer churn dataset, as well as in a combined multi-domain setting. Experimental results show that the framework achieves classification accuracies of up to 91.4% for employee attrition prediction and 87.3% for customer churn prediction, with corresponding ROC–AUC scores reaching 0.869 and 0.891, respectively. The model also demonstrates improved probability calibration, reducing the Brier score to 0.128 on the attrition dataset and 0.109 on the churn dataset. Furthermore, explainability analysis yields higher SHAP fidelity scores exceeding 0.90, while the counterfactual module generates valid and plausible recommendations with lower action costs compared to baseline approaches. These results indicate that the proposed framework effectively bridges predictive performance, interpretability, and actionability, making it suitable for real-world managerial decision-support systems. The framework is formulated as a constraint-aware optimization problem that integrates predictive modeling, explanation fidelity, and cost-sensitive counterfactual generation into a unified objective, enabling actionable and feasible decision support.
Related
- An integrated fuzzy multi-criteria decision making and mixed-integer linear programming optimization framework for project risk assessment and response strategy selection: A case study of a liquefied petroleum gas pipeline construction project
- Explainable fine-grained feature enhancement framework for strawberry diagnostic decision making
- scDEFT: A deep learning framework for drug-effect prediction and counterfactual reasoning
- LeafXNet: A Lightweight Explainable Model With Integrated Attention Mechanism for Fruit Leaf Disease Detection
- 64 Explainable AI for objective immunohistochemistry antibody clone selection, failure attribution, and optimization
BibTeX
@article{jahan2026integrated,
title = {{An integrated explainable AI framework with counterfactual optimization for decision support}},
author = {Nusrat Jahan and Jannatul Maua and Uland Rozario and Hashibul Ahsan Shoaib},
journal = {PLOS One},
year = {2026},
month = sep,
doi = {10.1371/journal.pone.0358643},
url = {https://doi.org/10.1371/journal.pone.0358643},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - An integrated explainable AI framework with counterfactual optimization for decision support
AU - Nusrat Jahan
AU - Jannatul Maua
AU - Uland Rozario
AU - Hashibul Ahsan Shoaib
PY - 2026
DA - 2026/09/23
JO - PLOS One
DO - 10.1371/journal.pone.0358643
UR - https://doi.org/10.1371/journal.pone.0358643
AB - Artificial intelligence is increasingly used to support managerial decision-making in domains such as human resource management and customer relationship management; however, the lack of transparency in complex predictive models limits their practical adoption. This paper proposes a counterfactual-guided explainable artificial intelligence framework that integrates deep learning, feature attribution, and counterfactual reasoning to deliver accurate, interpretable, and actionable managerial insights. The proposed model is evaluated on an employee attrition dataset and a customer churn dataset, as well as in a combined multi-domain setting. Experimental results show that the framework achieves classification accuracies of up to 91.4% for employee attrition prediction and 87.3% for customer churn prediction, with corresponding ROC–AUC scores reaching 0.869 and 0.891, respectively. The model also demonstrates improved probability calibration, reducing the Brier score to 0.128 on the attrition dataset and 0.109 on the churn dataset. Furthermore, explainability analysis yields higher SHAP fidelity scores exceeding 0.90, while the counterfactual module generates valid and plausible recommendations with lower action costs compared to baseline approaches. These results indicate that the proposed framework effectively bridges predictive performance, interpretability, and actionability, making it suitable for real-world managerial decision-support systems. The framework is formulated as a constraint-aware optimization problem that integrates predictive modeling, explanation fidelity, and cost-sensitive counterfactual generation into a unified objective, enabling actionable and feasible decision support.
SN - 1932-6203
ER -