Development of Pavement Condition Index Modelling Using Machine Learning Techniques

Résumé

The Pavement performance modelling is critical for sustainable transportation infrastructure,
since deterioration caused by traffic loads, weather conditions, and structural distress has a direct impact
on serviceability and maintenance costs. This research compared two prediction approaches for predicting
Pavement Condition Index (PCI): Artificial Neural Network (ANN) and Random Forest (RF) using key
deterioration variables: patches, potholes, temperature, depressions and cumulative Equivalent Single Axle
Loads (ESAL). The model demonstrated strong predictive performance, with RF achieving R² = 0.916 (RMSE
= 3.42), and ANN attaining the highest precision with R² = 0.961 (RMSE = 2.34). Sensitivity analysis
revealed that temperature, traffic loading, and potholes were the most important indicators, with patching
and depressions having little significance. The ANN model improved better predictive capability and RF
balanced accuracy with shifting importance. Collectively, technologies enhance improve data-driven pavement
management by allowing for accurate PCI forecasting and permitting proactive, cost-effective maintenance
planning for resilient transportation networks