Mapping at-risk transportation infrastructure assets using statistical and machine learning methods.

Area of Science:

• Geotechnical engineering
• Transportation infrastructure management
• Geographic Information Systems (GIS)

Background:

• Highway embankments and slopes (HWS) are vital but often overlooked transportation assets.
• HWS are susceptible to landslides, exacerbated by extreme rainfall events.
• Accurate mapping of vulnerable HWS is essential for effective infrastructure management.

Purpose of the Study:

• To develop and evaluate machine learning models for mapping at-risk highway embankments and slopes.
• To create a reliable inventory of vulnerable HWS for proactive asset management.
• To identify key factors influencing HWS failures.

Main Methods:

• Utilized Digital Elevation Models (DEMs) from remote sensing data to derive causative factors.
• Developed supervised machine learning models including Random Forest.
• Trained models using geotechnical, geomorphological, and hydrological data, with known HWS failure locations as ground truth.
• Evaluated model performance using AUC, F1-score, and Accuracy metrics.

Main Results:

• Random Forest model achieved perfect scores (AUC, F1, Accuracy = 1.0).
• An optimal probability threshold of 0.75 was determined to balance prediction accuracy.
• Key factors influencing HWS failures identified as elevation, distance from streams, NDVI, and precipitation.

Conclusions:

• The developed GIS-based machine learning method effectively maps vulnerable HWS across large areas.
• This approach enables targeted interventions and optimized fund utilization for infrastructure maintenance.
• Transportation agencies can adopt this methodology for strategic geotechnical asset management.