A Novel Incentive Mechanism Based on One-to-Many Bargaining in Coalition Federated Edge Learning for MEC-Enabled Consumer Electronics

Intelligent consumer electronics—smartphones, wearables, and IoT gateways—generate massive personal data at the network edge, driving demand for privacy-preserving collaborative intelligence. Federated edge learning (FEL) integrates federated learning (FL) with mobile edge computing (MEC) to enable distributed model training without exposing raw data. However, centralized FEL imposes prohibitive communication overhead when users are dispersed across multiple edge servers, while decentralized FEL suffers from degraded model accuracy due to limited local aggregation. Both paradigms ignore the fundamental asymmetry between mobile devices, which are resource-limited and intermittently available, and edge servers (ESs), which are fixed and resource-rich. To address this limitation, we propose coalition federated edge learning (CFEL), a framework that partitions users into ES-led coalitions where each ES serves as the coalition leader responsible for model aggregation and payment distribution. We model the incentive mechanism within each coalition as a one-to-many concurrent bargaining game (OCBG) and derive a closed-form Nash Bargaining Solution (NBS) that yields a principled, equal-surplus-sharing payment allocation grounded in MEC-specific communication and aggregation costs. Federated learning experiments on Fashion-MNIST and UCI-HAR demonstrate that CFEL achieves 89.24% accuracy on Fashion-MNIST (vs. 86.47% for decentralized FEL) and 90.80% on UCI-HAR (vs. 84.47% for decentralized FEL), while reducing distance-weighted communication cost by 68% compared with centralized FEL. Under non-IID data, periodic cross-coalition synchronization enables CFEL to surpass baselines on both datasets (82.66% on HAR, 81.44% on Fashion-MNIST). NBS payment allocation is further verified feasible and incentive-compatible under realistic MEC cost parameters. These results demonstrate that CFEL offers a favorable accuracy–communication trade-off for resource-constrained consumer electronics deployments. Source code is available at https://github.com/hl5w/CoalitionFederated-Edge-Learning.