Anomaly detection in IoT : federated learning approach on the IoT-23 Dataset.

The use of smart devices with software, sensors and other technologies(IoT) has become more common in the last few years. They aim to be interconnected through a network and exchange data with the purpose of reducing human intervention over specific tasks. This increase in IoT devices has incremented the data transit over a network. These devices' interactions and structure have raised the problem of malware attacks, especially over these devices, which can be manipulated to replicate more attacks. Because of this, there is a concern for identifying potential attacks. AI appears as a solution to this problem, and the Federated Learning approach was used as a technique of Machine Learning and using the IoT23 as Data-set to train and test the model. A Multi-layer Perceptron was used for both central server and federated members and Federated Averaging as an aggregation technique. The results showed that Federated Learning has the potential as a decentralized technique, which means that it benefice each device in its inner ML model. The model scored 100\% accuracy on some devices. However, the general learning was affected by this decentralized method having at most 70\% of accuracy. The model's performance was affected because of the data, distribution and quantity of each member's dataset.