du-CBA: Data-agnostic and incremental classification-based association rules extraction architecture

Abstract

In this study, federated learning architecture is used for training machine learning models without sending raw data from clients to the server in systems where clients and servers need to work together. According to the architecture, a machine learning model is trained on each client from its own data. The trained model is sent to the server and a new model is created by merging these models on the server. The final model created is distributed to the clients again. In order to realize the proposed architecture in the simulation, an algorithm called Data Unaware Classification Based on Association (du-CBA) has been developed. Experimental results showed that the model training time was reduced by approximately 70% with du-CBA compared to CBA, providing almost the same accuracy. The working logic of the federated learning architecture is shown in Figure A. Figure A. The working structure of the federated learning architecture Purpose: The aim of the study is to create an up-to-date model that provides data privacy by using machine learning methods for edge devices (phones, computers, IoT devices, etc.) whose place in our lives is gradually increasing.Theory and Methods: The du-CBA algorithm was developed in the simulation environment to implement the federated learning architecture for the study. The model consisting of labeled association rules is trained with the algorithm. Support and trust parameters are used when creating rules. The process of combining the models in the algorithm is realized by updating the support and confidence values of the rules and reordering them. Formulas have been developed to update support and trust values.Results: As a result of the experiments, it has been shown that the du-CBA algorithm developed for the federated learning architecture reduces the model training time by approximately 70% compared to the CBA algorithm and achieves almost the same accuracy. These results show that the proposed architecture has been successful. Conclusion: The study shows that with the federated learning architecture, network traffic is reduced, energy needs are reduced, and data privacy is protected because meaningless data is sent instead of all data.