Journal of Materials and Engineering Structures
Volume 13, Numéro 3, Pages 465-479
2026-09-30

Comparative Analysis Of Isolation Forest And One-class Svm For Outlier Detection In Gnss-rtk Cable-stayed Bridge Monitoring Data



Authors : Le Khanh Giang .

Abstract

Accurate GNSS-RTK displacement data are essential for reliable structural health monitoring (SHM) of long-span cable-stayed bridges; however, these measurements are often corrupted by noise, multipath, and gross errors. This study conducts a comparative evaluation of two unsupervised anomaly detection algorithms—Isolation Forest (IF) and One-Class Support Vector Machine (OC-SVM)—using a dataset of 2,078 GNSS-RTK displacement samples collected at the midpoint of the Can Tho cable-stayed bridge. The results show that both algorithms successfully identified major abnormal measurements with 70–73% overlap in detected outliers across the X, Y, and Z axes. IF detected 101–104 outliers, while OC-SVM detected 98–105, depending on the axis. IF consistently achieved higher cluster separability, with Silhouette Scores of 0.678 (X), 0.571 (Y), and 0.453 (Z), compared with 0.672, 0.563, and 0.433 for OC-SVM. Moreover, IF significantly outperformed OC-SVM in reconstruction fidelity, yielding Reconstruction Errors of 0.600–0.839, markedly lower than the 3.060–3.914 obtained by OC-SVM. After filtering out the detected anomalies, the GNSS-RTK displacement series showed substantial improvements in data stability; the standard deviation was reduced by 44.24% (X), 18.90% (Y), and 12.08% (Z) for IF and by 43.98% (X), 17.98% (Y), and 10.51% (Z) for OC-SVM. These quantitative findings demonstrate that both methods enhance GNSS-RTK data quality, while the Isolation Forest algorithm offers superior robustness, better cluster discrimination, and improved preservation of true structural responses. The results confirm the suitability of IF as a reliable, computationally efficient preprocessing tool for real-time GNSS-based SHM of cable-stayed bridges.

Keywords

GNSS-RTK, Outliers ; Isolation Forest ; One-Class SVM ; Displacement Monitoring ; Cable-Stayed Bridges

Les articles similaires

Study To Process Abnormal Data For Gnss Monitoring System Of A Long-span Cable-stayed Bridge In Vietnam

Le Van Hien . 
pages 421-426.

Deep Convolutional Neural Network-based Transfer Learning Method For Health Condition Identification Of Cable In Cable-stayed Bridge

Tran Viet-linh . 
pages 5-18.

Damage Detection For A Cable-stayed Bridge Under The Effect Of Moving Loads Using Transmissibility And Artificial Neural Network

Bui-tien Thanh .  Le-xuan Thang .  Tran-viet Hung .  Nguyen-xuan Tung .  Nguyen-ngoc Long .  Dao-thuy Linh .  Tran-ngoc Hoa . 
pages 411-420.

Evaluation Of The Static Behavior Of Cable-stayed Bridge With A Curved Pylon: A Case Study

Dinh Nguyen Duc Thi Thu .  Mac Van Ha . 
pages 345-359.

Wind Induced Vibration Of Stay Cable Bridge Evaluation Based On The Operational Accelerometers Monitoring Data And Field Testing

Phuong Duy Nguyen .  Lan Tran Duc . 
pages 597-603.

Formats de citation