# A Machine Learning Surrogate for Component Criticality Ranking in Interdependent Power-Communication Networks

arXiv: https://arxiv.org/abs/2607.08918 · HTML: https://mlsanity.com/paper/2607.08918/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 3 of its 12 references that are in the map.
- First posted on arXiv 9 Jul 2026.

## Builds on (most cited first, 3 of 3)

- [A New Model to Analyze Power and Communication System Intra-and-Inter Dependencies](https://mlsanity.com/paper/2001.06610/) (2020)
- [Identification of the K-most Vulnerable Entities in a Smart Grid System](https://mlsanity.com/paper/2008.08074/) (2020)
- [A Self-Updating K-Contingency List for Smart Grid System](https://mlsanity.com/paper/2101.08896/) (2021)

## Abstract (arXiv)

Cyber-physical power systems are vulnerable to cascading failures caused by interdependencies between power and communication infrastructures. Because evaluating large N-k contingency sets with a high-fidelity simulator is computationally expensive, this paper develops a machine-learning surrogate using the previously published Modified Implicative Interdependency Model (MIIM) as the ground-truth cascade simulator. The surrogate predicts contingency severity from leakage-free structural features and derives an association-based component-criticality ranking for resilience screening. On the IEEE 118-bus system, Gradient Boosting achieves a held-out Spearman correlation of 0.849 for contingency-severity ranking. Using five-fold out-of-fold predictions, the resulting component ranking achieves a Spearman correlation of 0.838 with the MIIM-derived ranking and closely approaches the observed cross-sample reproducibility level. Feature-ablation results show that inter-layer dependency features drive most of the surrogate's advantage, while end-to-end screening is approximately 158x faster than direct MIIM evaluation. The results support a two-stage workflow in which the surrogate screens candidate contingencies and components, and MIIM provides selective verification rather than directly identifying optimal hardening actions.