غزاله شيرواني

The Internet of Things (IoT) has become one of the fundamental elements shaping today's computer networks. The scope of IoT utilization ranges from everyday applications to sensitive domains within organizations and security matters. In this scenario, one of the most significant challenges is achieving an acceptable level of security. Distributed Denial of Service (DDoS) attacks, as a preva‎lent threat in this field, present a challenging security issue for IoT devices. These attacks not only threaten the device information but also the security of other devices connected to the network. Given the scale and complexity of destructive attacks and the evolving tactics of attackers, detecting and mitigating these threats on limited IoT devices pose intricate challenges. In this regard, one effective solution involves the use of machine learning techniques, particularly distributed approaches. However, due to the sensitivity of their data, many organizations cannot send their data beyond their boundaries for training machine learning models. This necessitates innovative solutions like federated learning. Federated learning enables the creation of a collaborative and robust machine learning model without the need to share data among multiple participants. This approach aids in achieving important goals such as data privacy, data security, and heterogeneous data management. This research aims to bridge the gap between the current state of federated artificial intelligence and its broader future acceptance. The proposed solution in this study employs federated learning and deep autoencoders to detect DDoS attacks using the N-BaIoT dataset. The model is trained and eva‎luated using the FedAvg and FedAvgM algorithms, with results demonstrating that the FedAvgM model outperforms with a lower false-positive rate. At the study's conclusion, the proposed model is compared with a corresponding traditional learning model (deep autoencoder). The comparison results reveal that the federated model exhibits superior performance in terms of detection capability and temporal efficiency, making it highly suitable for real-time applications.

اينترنت اشياء , يادگيري ماشين , يادگيري مشاركتي(فدرال) , امنيت شبكه , حملات DDoS

Internet of Things , Machine Learning , Federated Learning , Network Security , DDoS Attacks

Ghazaleh Shirvani

Mohammad Abdollahi Azgomi