Abstract
To address the frequent occurrence of sinkhole diseases in loess railway subgrades in Northern Shaanxi, China, this study systematically investigates their formation mechanisms and key influencing factors, provides a scientific basis for hazard prevention. Through field investigations of 134 sinkhole cases along the Baotou-Xi’an Railway, we established, incorporating five influencing factors: topography, hydrological conditions, design standards, construction grades, and subgrade types. A Bayesian network-based probabilistic prediction model was developed. The study reveals that complex topography (66.7% hazard incidence) and hydrological conditions (79.2% incidence) are the primary triggers, with continuous rainfall exceeding 50 mm resulting in an 81.5% hazard probability. High-fill subgrades, accounting for 86.57% of cases due to weak erosion resistance, Additionally, insufficient design and construction standards significantly increase risks, while improved engineering specifications can reduce hazard probability. Compared to traditional empirical methods, the model demonstrates enhanced objectivity and dynamic adaptability, fills the gap in systematic prediction of loess railway subgrade sinkholes, and provides a quantifiable decision-making tool for engineering practices.