Federated Learning with Adaptive Deep Rule-Based Classifier

Abstract

In the past, data processing required the transportation of data to computational re- sources. However, the current paradigm entails the deployment of computational resources to the data’s location. Federated Learning, a specialized domain within machine learning, employs a decentralized methodology for model training. Rather than centralizing the training process on a single server, this method allows for the local training of classifiers on individual devices, with model adjustments or parameter updates being disseminated to a central repository. This manuscript outlines two distinctive methodologies that are intended for implementation within Federated Learning frameworks. The first methodol- ogy integrates a scalable Decision Tree algorithm within a Federated Learning context to introduce a privacy-conscious mining strategy for large personalized data. This approach employs a Decision Tree classifier within the RainForest framework, guaranteeing that the central server does not receive any personal information. Each participating device trains the Decision Tree classifier locally and transmits their derived probability values as parameters to the central server. Using five benchmark datasets, we assessed the ef- ficacy of this methodology, which demonstrated exceptional performance. The following methodology involves the integration of traditional rule-based classifiers into a Federated Learning environment. This method, which is based on supervised learning, employs a set of IF-THEN rules that are derived by a Decision Tree classifier. This method gener- ates classification rules by utilizing local datasets on client devices. The rules are then transmitted to the central server for examination and integration. The iterative process enables the continuous refinement and adaptation of classification rules in response to the evolution of concepts. The efficacy of this method was validated through experimentation with ten benchmark datasets, which illustrated remarkable performance.