Over the years, LoRaWAN has enabled massive Internet of Things deployments by providing long-range, low-power connectivity for applications such as smart metering, environmental monitoring, and industrial sensing. Well-known protocol limitations, such as duty-cycle constraints and restricted capacity, have been extensively studied in the literature, together with various proposals aimed at improving scalability and reliability. More recently, commercial deployments have reached unprecedented scales, with tens or even hundreds of thousands of devices simultaneously, that make new challenges emerge. In particular, large geographical dispersion and logistical constraints prevent operators from reacting immediately to network anomalies. In this work, we present a data-driven diagnostic framework designed to support network operators in the maintenance and operation of large-scale LoRaWAN infrastructures. The proposed model uses protocol-aware multi-layer feature engineering to characterize device behavior across physical, data, and network dimensions, and combines unsupervised behavioral clustering with probabilistic health scoring and temporal forecasting of aggregated reliability indicators, enabling the identification and predictive monitoring of correlated degradation patterns. The model is validated on a real industrial deployment of more than 200,000 smart meters in central Italy. We show that the proposed framework assists operators in diagnosing network anomalies and reveals geographically correlated clusters of degraded devices, thereby enabling targeted and cost-efficient maintenance interventions.
Casalbore et al. (Fri,) studied this question.