Application of Federated Learning to Detect Malicious Behavior in Internet of Vehicles

Abstract

With the escalating application of the Internet of Things (IoT) in several fields, implementation of IoT in the automotive ecosystem is currently one of the critical concerns due to the enormous potential for its expansion in unimaginable ways. Internet of Vehicles (IoV) applies to the present-day human-driven vehicles as well as impending autonomous ones. Smart transportation is significantly safer, cost-effective, more convenient, and more efficient. Despite offering plenty of benefits, IoV face serious issues including big data problems, user security and privacy, and vehicle reliability. Reliable connection channels are established but this doesn't eliminate the cyber risks associated with them. With the increasing frequency of these security incidents in IoV, guarding against these attacks has been the foremost priority. Regardless of the standard protocols and established frameworks, these attacks are still likely to endanger the vehicle and user privacy and security. To address the security and privacy issues, the primary focus of this paper is the application of federated learning to detect attacks on the security and privacy aspects of the IoV. Without using centralized data, the federated learning technique develops the prediction model utilizing user data from the devices. Thus, the model is collaborative and shared, and as model training comes down to devices, the user's data is secure as training data resides on the device and no specific versions are maintained in the cloud. Hence, the main objective is to employ a federated learning approach to ascertain any kind of malevolent conduct in the connected vehicle systems and propagate trusted, authentic and reliable information for better deployment.