ISSN (Print) : 2320 – 3765 ISSN (Online): 2278 – 8875 International Journal of Advanced Research in Electrical, Electronics and Instrumentation Engineering (A High Impact Factor, Monthly, Peer Reviewed Journal) |Impact Factor: 6.392| Website: www.ijareeie.com Vol. 7, Issue 1, January 2018 Copyright to IJAREEIE DOI:10.15662/IJAREEIE.2018.0701001 1 Federated Learning: An Intrusion Detection Privacy-Preserving Approach to Decentralized AI Model Training for IOT Security Mohit Mittal Dr. A.P.J. Abdul Kalam Technical University, Lucknow, Uttar Pradesh, India ABSTRACT: There are various aspects to Internet of Things security, such as guaranteeing the safety of both the devices and the Internet of Things networks to which they connect. Many other types of equipment, including industrial robots, smart grids, construction automation systems, entertainment gadgets, and many more, are included in this, despite the fact that they were not designed with network security in mind. When it comes to securing systems, networks, and data, IoT device security must be able to resist a wide range of IoT security assaults. One of the most important issues in the field of data security is the creation of intrusion detection systems (IDSs) for the Internet of Things. Client devices (edge devices) in Federated Learning utilize local data to train the machine learning model, and then send the updated model parameters to a cloud server so that they may be aggregated (rather than raw data). This paper proposes a machine learning system that employs federated learning to detect intrusions in the IoT. The FedAVG algorithm is employed to aggregate models. Models are trained locally by nodes. The models are trained and validated using machine learning methods, including Random Forest, ID3, and Support Vector Machine. The NSL KDD data set is employed to undertake experiments. KEYWORDS: Federated Learning, Model Aggregation, TensorFlow, Machine Learning, Intrusion Detection, Internet of Things, Accuracy, FedAVG I. INTRODUCTION A centralized server is essential for Federated Learning (FL) to consolidate parameters. Central server malfunctions may result in single points of failure (SPoFs) and distributed denial of service (DDoS) attacks. Local model changes must be explicitly documented by FL. The capacity of a distributed system to detect and avert illegal modifications is essential to its efficacy. Ensure the security of FL systems by the use of blockchain technology. Federated deep learning and blockchain are used to mitigate the duration of poisoning and assaults. This model is designed for efficiency and performance in decentralized data training. Modifications to machine learning models may be safely and openly documented via distributed ledgers. Due to blockchain's capacity to prevent retractions, Federated Learning can more effectively identify collaborative machine learning models and enhance system trust. This work focusses on the audibility of blockchain-federated deep learning. At the conclusion of each round, the global model is synthesized from the local models of all Federated Learning participants by Federated Averaging (FedAvg). All local models get advantages from FedAvg averaging. The inadequacy of the local model in traditional FedAvg always results in the global model's failure. Dynamic weighted updating The Internet of Things (IoT) is a network that connects a broad variety of machines, devices, and other items that are connected to the internet through the use of wireless communication technology. A portion of the home is outfitted with hardware from the Internet of Things, which is also used in said location. An attack on the Internet or any of the systems linked with it has a substantial influence on the operating environment of the internet of things (IoT). A wide range of sensor nodes and devices, each of which makes use of a different technology, may be found at each of the Internet of Things' layers. IPv4 has a finite number of accessible locations, which is why Internet of Things devices utilize IPv6 rather than the older protocol. It is conceivable for Internet of Things devices to have smart meters, sensors,