APPLICATION OF EXPLAINABLE ARTIFICIAL INTELLIGENCE IN GLASS IDENTIFICATION BASED ON ITS CHEMICAL COMPOSITION

Abstract

The identification of glass based on its chemical component composition is frequently encountered in scientific literature. However, in most cases, users and even experts cannot understand the underlying reasoning behind the results obtained from machine learning algorithms and black box models. Despite the existence of numerous explainable systems, the problem of black box model explainability remains perpetually relevant. This paper presents a novel approach for providing explanations based on the classification of substances according to their chemical composition. Our proposed idea is based on DARPA's explainable artificial intelligence architecture. This architecture was implemented by Bellucci et al., who extended it with semantic web technologies in their work. The ontological properties proposed by Kosov P.İ. et al. are applied in this paper. The explainable system developed in this work is specifically designed for the explainability of chemical components and is based on relevant expert knowledge presented as ontology. The designed system has been verified and tested through experimentation on a glass identification dataset containing chemical properties and individual elements in the description of objects.