Dispersed Data Classification Model with Conflict Analysis and Parameterized Allied Relations
Małgorzata Przybyła‐Kasperek, Katarzyna Kusztal, Benjamin Agyare Addo · Procedia Computer Science · 2024
In the paper, a classification model for dispersed data is proposed. By dispersed data we mean a set of local tables that are collected independently by different units. This model uses conflict analysis, which considers the similarity of the values of conditional attributes occurring in decision classes of local tables. Tables having compatible values within classes are arranged into coalitions. The paper proposes the use of a parameter that steers the intensity of conflict between tables that are in allied relation. It was experimentally confirmed using 25 dispersed data that the proposed approach gives better results than the baseline approach in which coalitions are not used. The statistical significance of the differences was also proven. In addition, an advanced analysis of the impact of the parameter’s value steering the allied relations on the form of coalitions and the quality of classification was carried out.