Computer science and information processes

Development of an algorithm based on logical operations to detect patterns in data with missing values

L. A. Lyutikova

Institute of Applied Mathematics and Automation – branch of Kabardino-Balkarian Scientific Center of the Russian Academy of Sciences, 360000, Russia, Nalchik, 89 A Shortanov street

Abstract: This paper presents a method of local interpretation of solutions of a trained neural network by functions of multivalued predicate logic. The local interpretation of a neural network refers to the process of explaining the decisions made by the model on a specific example or in the vicinity of a specific input. The proposed approach is based on a set of functions of multivalued logic, which are generalized operations that meet certain requirements. By combining these functions, it is possible to detect internal patterns in the data and even correct the results obtained with the help of neural networks. The proposed method was investigated in the context of classification problems using multidimensional discrete features. In such cases, each attribute can take one of k possible values and have equal importance for class identification. This approach opens up new possibilities for understanding and explaining the rules underlying the data, which are not always obvious when using conventional neural networks.

Keywords: interpretation, multivalued logic, neural network, generalized addition, data

UDC: 004.83

MSC: 68T27

Received: 30.11.2023
Revised: 01.12.2023
Accepted: 10.12.2023

DOI: 10.35330/1991-6639-2023-6-116-109-115

Bibliographic databases:

•