Interval-valued fuzzy predicates from labeled data: An approach to data classification and knowledge discovery

Abstract

Interpretable data classifiers play a significant role in providing transparency in the decision-making process by ensuring accountability and auditability, enhancing model understanding, and extracting new information that expands the field of knowledge in a discipline while effectively handling large datasets. This paper introduces the Type-2 Label-based Fuzzy Predicate Classification (T2-LFPC) method, in which interval-valued fuzzy predicates are used for interpretable data classification. The proposed approach begins by clustering the data within each class, associating clusters with collections of common attributes, and identifying class prototypes. Interval-valued membership functions and predicates are then derived from these prototypes, leading to the creation of an interpretable classifier. Empirical evaluations on 14 datasets, both public and synthetic, are presented to demonstrate the superior performance of T2-LFPC based on the accuracy and Jaccard index. The proposed method enables linguistic descriptions of classes, insight into attribute semantics, class property definitions, and an understanding of data space partitioning. This innovative approach enhances knowledge discovery by addressing the challenges posed by the complexity and size of modern datasets.