Relationships between attributes and decisions with exclusion of part of the data using fuzzy inequalities

The study of relationships between data aims to simplify the models and provide explanations for medical, social, financial and other phenomena. This article introduces an approach to analyzing the relationships between attributes and decisions using relational systems of inequalities, incorporating a certain degree of trust in the data. During analyzing real data, a subset of results for a certain attribute often deviates from the rest of the examined group. Therefore, separate analyses are worth considering for such groups. The proposed method enables the exclusion of a specified subset of results and determines the highest degree of dependence between the examined attributes and the decision. The excluded data are stored in separate files, allowing for further analysis in subsequent stages. However, for many researchers, mathematical models are difficult to apply due to their dependence on advanced mathematical concepts or high computational complexity. Additionally, the consequence of this work is an application RRADD (Reduction-Relation-Attributes-Decision-Data) that enables the exploration of relationships based on this method, facilitating practical implementations. In this research, the proposed RADEPD method is applied to the BostonHousing data set. The example demonstrates how separating outlier observations improves the interpretability of relationships between attributes and the decision, and how it affects classification accuracy under different t-norms.