FedGATE: A Reputation-Aware Federated Learning Framework for Robust Healthcare Fraud Detection under Heterogeneous Institutional Data

Cross-institutional healthcare fraud detection is constrained by privacy regulations that prevent medical and insurance organizations from pooling raw claim records. Federated learning (FL) enables collaborative training without exchanging raw healthcare records, but practical deployments remain vulnerable to non-independent and identically distributed (non-IID) data, extreme class imbalance, volatile participation, and under-trained local updates. This paper presents FedGATE, a reputation-aware FL framework for healthcare fraud detection across heterogeneous institutions, with an emphasis on reducing the influence of under-trained client updates. FedGATE evaluates each local update through directional consistency, historical update stability, and task effectiveness. It recursively accumulates these signals into long-term reputation values and maps them to dynamic aggregation weights, thereby strengthening useful updates while limiting the influence of unreliable ones. Experiments on two real-world healthcare fraud datasets show competitive performance on the public Healthcare Provider Fraud Detection Analysis (HPFDA) dataset. On the more heterogeneous anonymized real-world healthcare settlement dataset (hereafter referred to as XNYB), FedGATE improves area under the precision–recall curve (AUPRC) by 4.28 percentage points over Federated Averaging (FedAvg). When the proportion of under-trained clients increases from 20% to 60%, FedGATE maintains an AUPRC of approximately 0.65 with a range of only 0.0036, and retains the highest average AUPRC even when under-trained clients form the majority of selected participants. Ablation studies further verify the roles of multidimensional contribution assessment, dynamic weighting, and client participation. These results demonstrate that FedGATE provides a lightweight and deployment-oriented aggregation layer for reliable federated healthcare fraud analytics.