EUVIMEDEUVIMED entdecken
Crossref

Method for constructing an explanation tree with a compact structure for interpreting intelligent-system decisions

S.F. Chalyi, A.M. Dementiev

Radiotekhnika · 30.09.2026 · (226) · 7-16

DOI: 10.30837/rt.2026.3.226.01 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Radiotekhnika, online erschienen am 30.09.2026. Verfasst von S.F. Chalyi und 1 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Intelligent information systems are widely used to solve problems related to decision-making concerning human health and well-being. In such tasks, it is insufficient for a user to receive only a class label, a prediction, or a recommended course of action. The conditions that determined the formation of an intelligent-system decision should also be presented. This problem is particularly important when opaque machine-learning models are used because the internal logic of their operation cannot be directly interpreted by users. The object of the study is the process of constructing explanation trees for interpreting a set of similar decisions made by an intelligent system. The aim of the study is to develop a method for constructing an explanation tree with a compact structure for interpreting a set of similar intelligent-system decisions. To achieve this aim, the following tasks are addressed: developing an approach to constructing explanation trees for a set of similar intelligent-system decisions; developing a method for constructing an explanation tree with a compact structure; and experimentally evaluating the proposed method in the task of constructing explanations. An approach to constructing an explanation tree is proposed, in which decisions of the base intelligent system are represented through refinement conditions and transitions between them. The approach enables the construction of a generalized explanation for a group of objects with similar decisions made by the base system. A method for constructing an explanation tree with a compact structure is developed. The method involves generating alternative trees, determining their fidelity to the decisions of the base model, evaluating structural complexity, and selecting the resulting tree. Alternatives that reach a specified fidelity threshold with respect to the decisions of the base intelligent system are included in further consideration. Among admissible alternatives, a tree with fewer structural elements is selected. An experimental evaluation of the method was conducted using a random forest as the base model. The results confirmed that the proposed method enables the selection of a compact explanation tree that preserves fidelity to the decisions of the base model. The constructed tree is a surrogate model that reproduces the behaviour of the base system. Its conditions do not establish causal mechanisms in the application domain. Further development of the method is related to examining the possibility of constructing explanation trees in multiclass classification tasks and to evaluating the stability, understandability, and usefulness of the constructed explanations with user involvement.

Zitieren

APA

Chalyi, S., & Dementiev, A. (2026). Method for constructing an explanation tree with a compact structure for interpreting intelligent-system decisions. Radiotekhnika, 7-16. https://doi.org/10.30837/rt.2026.3.226.01

Vancouver

Chalyi S, Dementiev A. Method for constructing an explanation tree with a compact structure for interpreting intelligent-system decisions. Radiotekhnika. 2026:7-16. doi:10.30837/rt.2026.3.226.01

BibTeX
@article{Chalyi2026,
  author = {Chalyi, S.F. and Dementiev, A.M.},
  title = {Method for constructing an explanation tree with a compact structure for interpreting intelligent-system decisions},
  journal = {Radiotekhnika},
  year = {2026},
  number = {226},
  pages = {7--16},
  doi = {10.30837/rt.2026.3.226.01}
}
Kostenlos starten →