Federated Learning For Iot Forensics: Enabling Privacy-Preserving Machine Learning

IoT Forensics is an expanding research field that combines the use of IoT devices and their data to help solve crimes. Additionally, Machine Learning (ML) can be a valuable tool to assist investigators in solving crimes. However, the lack of shared data — particularly due to privacy concerns — can represent obstacles to the use of ML models and their fine-tuning for specific IoT Forensics scenarios. Due to intellectual property and confidentiality, most datasets are not shared and cannot be used to improve ML models. Therefore, the goal of this work is to propose a methodology that uses a Federated Learning (FL) approach in IoT Forensics contexts. This methodology is evaluated through the training of ML models without necessarily sharing user data. Consequently, data privacy can be ensured, as there is no need to share it with a central server, thereby reducing security and confidentiality risks. The evaluation shows that the proposed methodology achieved a 99.4% Area Under the Curve (AUC) score in the global model trained using a FL approach. Finally, our proposed methodology can support the adoption of FL in IoT Forensics, keeping users' data private.