A model-agnostic and data-independent tabu search algorithm to generate counterfactuals for tabular, image, and text data

Raphael Mazzine Barbosa de Oliveira, Kenneth Sörensen and David Martens

European Journal of Operational Research, 2024, vol. 317, issue 2, 286-302

Abstract: The growing prevalence of artificial decision systems has prompted a keen interest in their efficiency, yet this progress is accompanied by their inherent complexity. This poses a significant challenge for various domains, including operational research, where decisions hold crucial influence over outcomes and thus must not remain undisclosed. Counterfactual explanations are greatly remarked as a simple (to understand) yet efficient way to explain the decisions made by a machine learning model by finding a minimal set of changes required to change the prediction outcome for a specific instance. We, then, present a novel algorithmic approach, called CFNOW, which implements a modular, fast, two-step process using tabu search, a well-known metaheuristic framework, to find counterfactuals for multiple data types (tabular, image, and text) with high efficiency. We run an extensive benchmark study with more than 5000 factual points from 25 datasets to demonstrate that CFNOW can generate high-quality counterfactual results in terms of metrics such as speed, coverage, distance, and sparsity, surpassing the state-of-the-art. These characteristics, associated with the simple code implementation, may aid embedding explainability to complex models which are often necessary for compliance requirements.

Keywords: Decision support; Counterfactual explanations; Explainable artificial intelligence; Deep learning (search for similar items in EconPapers)
Date: 2024
References: View references in EconPapers View complete reference list from CitEc
Citations:

Downloads: (external link)
http://www.sciencedirect.com/science/article/pii/S0377221723006598
Full text for ScienceDirect subscribers only

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:eee:ejores:v:317:y:2024:i:2:p:286-302

DOI: 10.1016/j.ejor.2023.08.031

Access Statistics for this article

European Journal of Operational Research is currently edited by Roman Slowinski, Jesus Artalejo, Jean-Charles. Billaut, Robert Dyson and Lorenzo Peccati

More articles in European Journal of Operational Research from Elsevier
Bibliographic data for series maintained by Catherine Liu ().