Cross-domain benchmarking of machine learning on internet of things sensor data for civil infrastructure monitoring: a unified dataset, leakage-aware evaluation, and standardisation gap analysis

The growth of the Internet of Things (IoT) in civil infrastructure has produced a heterogeneous knowledge base spanning structural health monitoring, water distribution, transportation infrastructure, geotechnical systems, construction site safety, and climate resilience. Despite considerable progress within each field, a shared analytical framework for comparing sensing architectures, machine learning applicability, network performance predictors, and standardisation deficits across these fields has remained absent. This study combines a systematic cross-domain review with a unified benchmark dataset of 550 sensor records, parameterised from published reference deployments across the six domains, and evaluates ten machine learning classifiers for binary alert detection under five-fold stratified cross-validation together with an independent 20% hold-out test set. To address the risk of label leakage through derived condition indices, all experiments were repeated under three feature configurations: the full feature set, a configuration excluding the composite Degradation Score, and a deployable-sensor-only configuration. Random Forest achieved the highest cross-validated accuracy of 0.8564 with the full feature set, declining to 0.8091 and 0.7909 under the two leakage-aware configurations, thereby quantifying the contribution of composite indices explicitly. A three-class severity extension (normal, advisory, critical) demonstrated the applicability of the framework to multi-class problems, with Gradient Boosting attaining a weighted F1-score of 0.6582 and a one-versus-rest AUC-ROC of 0.7892. Kruskal-Wallis tests with epsilon-squared effect sizes confirmed significant cross-domain differences in Technology Readiness Level, signal quality, and network reliability (p < 0.001), and post-hoc Dunn tests localised the network-quality differences predominantly to the Construction Site Safety domain. A standardisation gap analysis and a decision-centric research agenda translate these quantitative findings into actionable priorities for infrastructure owners, standards bodies, and the research community.