Decision rules for dispersed data using a federated learning approach

Abstract

The paper deals with dispersed data stored in independent local decision tables. We assume that there are the same conditional attributes in all tables. The paper proposes a new approach to generate global reducts based on data stored in all dispersed tables. The federated learning approach is used to ensure data protection and privacy. Based on the global reducts, decision rules are generated for each local table. Finally, the global classifier is composed of the set of all decision rules. In the paper, the proposed approach is compared with the baseline approach, in which the local reducts and local rules are generated for each individual table separately. The final decision is made using majority voting of the local models. It was shown that the proposed approach using federated learning provides better classification quality than the baseline approach.