Enhancing Privacy in IoT-Enabled Digital Infrastructure: Evaluating Federated Learning for Intrusion and Fraud Detection

. 2025 May 12;25(10):3043. doi: 10.3390/s25103043

Algorithm 2 FedProx: federated proximal algorithm.

1:
Input: Number of rounds T, number of clients K, learning rate $η$ , proximal term $μ$
2:
Initialize global model weights $w_{0}$
3:
for each round $t = 1, \dots, T$ do
4:
Sample a subset $S_{t}$ of K clients
5:
for each client $k \in S_{t}$ in parallel do
6:
Download global model weights $w_{t}$
7:
Compute local model weights $w_{t + 1}^{k}$ by performing E epochs of proximal SGD on local data, minimizing the objective: $L (w) + \frac{μ}{2} {∥ w - w_{t} ∥}^{2}$
8:
end for
9:
Aggregate local models: $w_{t + 1} = \frac{1}{K} \sum_{k \in S_{t}} w_{t + 1}^{k}$
10:
end for
11:
Return: final global model weights $w_{T}$