Explainability through uncertainty: Trustworthy decision-making with neural networks

• Published in: European journal of operational research Vol. 317; no. 2; pp. 330 - 340
• Main Authors: Thuy, Arthur, Benoit, Dries F.
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.09.2024
• Subjects: Decision support systems; Deep Ensembles; Distribution shift; Explainable artificial intelligence; Monte Carlo Dropout; Decision support systems; Explainable artificial intelligence; Monte Carlo Dropout; Deep Ensembles; Distribution shift
• ISSN: 0377-2217; 1872-6860
• DOI: 10.1016/j.ejor.2023.09.009

•Uncertainty is a key feature of any ML model, especially for overconfident NNs.•Uncertainty estimates indicate when an ML model’s output should (not) be trusted.•The proposed general uncertainty framework positions uncertainty as an XAI technique.•The framework uses classification with rejection, bringing an expert in the loop.•A case study in OR with distribution shifts applies the uncertainty framework to NNs. Uncertainty is a key feature of any machine learning model and is particularly important in neural networks, which tend to be overconfident. This overconfidence is worrying under distribution shifts, where the model performance silently degrades as the data distribution diverges from the training data distribution. Uncertainty estimation offers a solution to overconfident models, communicating when the output should (not) be trusted. Although methods for uncertainty estimation have been developed, they have not been explicitly linked to the field of explainable artificial intelligence (XAI). Furthermore, literature in operations research ignores the actionability component of uncertainty estimation and does not consider distribution shifts. This work proposes a general uncertainty framework, with contributions being threefold: (i) uncertainty estimation in ML models is positioned as an XAI technique, giving local and model-specific explanations; (ii) classification with rejection is used to reduce misclassifications by bringing a human expert in the loop for uncertain observations; (iii) the framework is applied to a case study on neural networks in educational data mining subject to distribution shifts. Uncertainty as XAI improves the model’s trustworthiness in downstream decision-making tasks, giving rise to more actionable and robust machine learning systems in operations research.