An Explainability Framework For Systems Fusing Base Black Boxes
Year: 2026
Type: conference-paper
Abstract: For the past two decades, there has been an increasing number of Decision Support Systems that draw their decisions on a number of underlying base algorithms that each provide their own recommendation to a higher-level fusion algorithm that makes the final recommendation. In Machine Learning related fields, this process is called ensemble fusion, and it works for almost all kind of problems: classification, regression, clustering and so on. A common characteristic trait of such ensembles is that the top-level fusion engine very often has little or no information as to how each base system arrived at its decision. However, today there is a fast growing need for Decision Support Systems to be able to explain their decisions. In this paper, we propose a modality-output-level explainability framework tailored to various fusion techniques (i.e., TOPSIS, Dempster–Shafer, Bayesian) with fusion under output-only constraints. Specifically, for the particular problem of entity (person) identification, we frame the problem of explaining the final decision an ensemble system makes in the context of various fusion methods and for one of them, we provide indicative results.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7204482060