FLVAMatch: an economic matching game theory-based vehicle selection in federated learning for the next-generation of internet of vehicles

The future generation of the internet of vehicles (IoV) in intelligent transportation systems (ITS) leverages advanced communication and intelligent data analysis. Federated learning (FL) enhances IoV privacy, but conventional random-based vehicle selection for local training is inefficient due to varying client resources and data quality. Existing methods prioritise server preferences, overlooking FL-vehicle incentives. To address this, we formulate a profit maximisation problem to enhance FL-vehicle revenues and propose FLVAMatch, an economic framework based on the hospital-resident (H-R) matching problem. In a three-layer software-defined vehicular fog (SDVF) computing setup, FLVAMatch considers both FL-vehicle and aggregator preferences. Simulation results demonstrate its effectiveness in maximising FL-vehicle revenues and reducing network latency, outperforming state-of-the-art FL approaches, with the global model achieving 88% accuracy.