Pavement condition prediction under small-sample conditions using a particle swarm optimization-based support vector...

Area of Science:

• Civil Engineering
• Transportation Engineering
• Data Science

Background:

• Effective pavement maintenance planning requires precise Pavement Condition Index (PCI) forecasting.
• Budget constraints necessitate efficient and accurate predictive models for infrastructure management.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting Pavement Condition Index (PCI).
• To compare the performance of Support Vector Machine (SVM), Back Propagation Neural Network (BPNN), and Particle Swarm Optimization-optimized SVM (PSO-SVM) for PCI forecasting.

Main Methods:

• Utilized field data from Chinese roads (ordinary and expressway) with five input factors: road age, traffic volume, temperature, precipitation, and humidity.
• Employed Particle Swarm Optimization (PSO) to tune hyperparameters (c and γ) for SVM models.
• Assessed model performance using a 70/30 hold-out split and 5-fold cross-validation for small-sample data.

Main Results:

• The PSO-SVM model demonstrated higher prediction accuracy and more stable performance than baseline SVM and BPNN models under small-sample conditions.
• Random Forest analysis identified road age, traffic volume, and temperature as key determinants influencing PCI.
• The optimized PSO-SVM model offers practical guidance for pavement repair decisions, particularly in frigid regions.

Conclusions:

• PSO-SVM is a highly effective method for Pavement Condition Index (PCI) forecasting, outperforming traditional models.
• Road characteristics like age, traffic, and climate significantly impact pavement condition.
• The developed models provide valuable tools for optimizing pavement maintenance strategies and resource allocation.