Network screening of derailment consequence severity using exposure proxies: Gyeongbu High-Speed Line
Jaehoon Lim,Seong Keun Kang,Kyumin Na,Hyunkyu Jun,Chanwoo Park,Jung Sik Kong
Abstract
This study presents a deterministic analytic protocol for consequence-severity screening that focuses on evaluating the behaviour of screening methods rather than on probabilistic risk estimation. The protocol represents escalation potential using interpretable structural context descriptors and geospatial exposure proxies, and evaluates screening outputs using rank-stability and screening-efficiency metrics. The analytic metrics include Spearman rank correlation, top-k overlap (Jaccard similarity), grade reclassification rates, and candidate inflation to quantify how modelling choices and baseline definitions expand or contract prioritised candidate sets. Corridor-scale decision utility is reported using contiguous high-grade clusters rather than isolated section scores. The protocol is demonstrated on the Gyeongbu High-Speed Line (Republic of Korea) under alternative segmentation resolutions (500 m and 1 km) and grading variants. Results show that combined structural-exposure screening yields more compact candidate sets than structural-only or exposure-only baselines while remaining robust to key analytical choices. It is intended for prioritisation and does not estimate derailment probability, calibrate losses, or constitute a full risk assessment.
