Abstract
<jats:p>Road surface condition monitoring is essential for enhancing transportation safety and infrastructure maintenance. This study develops an IoT-oriented inertial sensing framework for real-time pavement classification, pothole detection, and enhanced vehicle position estimation. The framework integrates a memory-constrained XGBoost model designed for microcontroller deployment, a velocity-aided GPS interpolation procedure, and an abnormality-index-based pothole detection algorithm. Experimental results on a private dataset and the PVS dataset show classification accuracies of 95.39% and 93.21%, respectively. To examine transferability, the configuration tuned on the private dataset was applied to the PVS dataset without retraining and achieved 92.45% accuracy. Furthermore, the GPS interpolation procedure reduces mean localization errors from 5.571–11.893 m to 1.835–3.563 m across vehicle speeds of 20–50 km/h. An additional contribution of this study is the release of a private dataset capturing vibration signatures from representative road types, supporting further research in road surface classification.</jats:p>