摘要
Assessing the reliability of large-scale binary-state networks is essential for resilient infrastructure, from power grids to communication systems. Exact computation is infeasible at scale, and conventional Monte Carlo simulations can be computationally expensive, particularly for high-reliability systems. This study presents an empirical evaluation of representative data-driven methods, including linear and polynomial regression, tree ensembles, support-vector regression, neural networks, and K nearest neighbors, across three reliability regimes: full range (0.0 1.0), high reliability (0.9-1.0), and ultra-high reliability (0.99-1.0). We identify a practical threshold: networks with arc reliability above 0.9 achieve near-unity system reliability, allowing significant simplifications in reliability assessment. Additionally, we establish guidelines for selecting approximation methods based on dataset size: neural networks perform effectively with moderate data volumes, while polynomial regression delivers higher accuracy when large training datasets are available. The findings provide actionable insights for balancing accuracy, computational cost, and interpretability in reliability evaluation, offering practical strategies for infrastructure analysis and system design. © 2025 IEEE.