HuddsTrafficFL: A benchmark testbed and dataset for Byzantine-robust federated learning in urban traffic systems

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Federated learning (FL) enables collaborative model training without sharing raw data, making it suitable for privacy-sensitive applications such as intelligent transportation systems and distributed Internet of Things (IoT) environments. However, evaluating Byzantine-resilient FL under realistic deployment conditions remains challenging. Existing benchmark datasets are mainly based on image classification or generic network intrusion tasks and often fail to capture geographically distributed clients, naturally occurring non-IID data, sensor-derived anomaly labels, and integrated Byzantine attack evaluation within a unified framework. To address this limitation, we present HuddsTrafficFL , a reproducible benchmark dataset and evaluation framework developed from a city-scale traffic sensing environment using SUMO/TraCI simulations and OpenStreetMap road networks. HuddsTrafficFL provides three core contributions: (1) a geographically grounded federated learning dataset containing 91,482 timestep observations collected from 292 clients, capturing natural non-IID characteristics and heterogeneous data distributions; (2) sensor-derived anomaly labels generated directly from junction-level traffic behaviour, including sudden speed reductions and abnormal queue growth patterns; and (3) a configurable Byzantine attack and defence evaluation pipeline that enables reproducible robustness assessment across multiple datasets. To the best of our knowledge, HuddsTrafficFL is the first benchmark that simultaneously integrates geographically grounded FL clients, sensor-derived anomaly labels, natural non-IID data characteristics, and Byzantine attack evaluation within a single reproducible environment. We evaluate four representative Byzantine attacks and three aggregation strategies: FedAvg (undefended baseline), Krum and Coordinate Median (robust methods) across HuddsTrafficFL, CICIDS2017, and BoT-IoT. Experimental results show that standard FedAvg achieves an AUC of 0.6706 under model poisoning with a 25% attacker fraction per round, whereas coordinate median consistently achieves an AUC of at least 0.9874 across all evaluated attack-dataset combinations. The baseline federated model on HuddsTrafficFL achieves an AUC of 0.9683, demonstrating that the benchmark provides realistic learning conditions and meaningful robustness evaluation. The dataset, simulation pipeline, interactive dashboard, and benchmark scripts will be publicly released to support reproducible research on secure federated learning for intelligent transportation systems and distributed sensing applications.