UDC 681.5
METHOD FOR IMPROVING THE EFFICIENCY OF INTELLIGENT DIGITAL TWIN MODELS BASED
ON ADDITIVE FEATURE ATTRIBUTION
V. P. Koryachko, Dr. in technical sciences, full professor, Head of CAD Department, RSREU, Ryazan, Russia;
orcid.org/0000-0003-0272-673X, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
M. S. Kosheleva, Postgraduate Student, RSREU, Ryazan, Russia;
orcid.org/0009-0000-5265-4733, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
V. I. Oreshkov, PhD (in technical sciences), Associate Professor, RSREU, Ryazan, Russia;
orcid.org/0000-0003-0316-4927, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
The article examines the problem of explainability of machine learning models used in intelligent digital
twins (IDTs). An analytical review of methods for enhancing the explainability of machine learning models
used in IDTs is provided in the context of explainability-accuracy dilemma. A new combined method for increasing the local explainability of features in a training dataset is proposed, based on the use of additive
attribution of features together with the assessment of predictive effectiveness of features on entire dataset.
Using the example of a neural network model for a digital twin of a credit manager, a numerical experiment
was conducted, which reflected the compliance of practical results with theoretical principles. Specifically,
the selection of features for training the model from the perspective of enhancing its explainability is shown
to be proved suboptimal in terms of its predictive effectiveness and, therefore, requires further verification.
The novelty of the work lies in the development of a methodology for combined use of additive attribution of
features together with the assessment of their predictive ability, which makes it possible to solve the problem
of explainability in the context of «explainability-accuracy» trade-off.
Key words: artificial intelligence, intelligent digital twin, explainable artificial intelligence, machine
learning models, self-explaining models, posthoc explanation, additive feature attribution method, explainability, interpretability.