Federated Isolation Forests for IIoT Anomaly Detection: Robust Aggregation and Poisoning Attack Mitigation

the work intends to apply a technique of tree-weight tagging and a variance based approach to the poisoning mitigation of a federated learning model used for anomaly detection: precisely isolation forest. Via a proper simulation of clients and a forest made of multiple trees, we managed to obtain a 35% improvement in precision of the model with respect to the base model.