<![CDATA[During road traffic operations, pavement quality directly affects safety, vehicle operating costs, and pavement maintenance activities. Traditional inspection methods are often costly and time-consuming, and they cannot provide continuous data on pavement conditions. This study aims to develop an efficient road-surface classification system capable of real-time operation on low-cost hardware devices. The system uses vibration data collected from vehicles in motion to identify and classify road types with high accuracy and optimized performance. The proposed system employs inertial sensors mounted on vehicles to acquire accelerometer and gyroscope signals and then extracts time-domain statistical features from these signals. To address the main challenge of deploying an effective recognition model in a resource-constrained computing environment, the paper proposes a hybrid feature selection algorithm that combines filter and wrapper methods. This algorithm leverages the fast-processing speed of filter methods and the effective feature selection capability of wrapper methods. The selected feature set is then evaluated using three machine learning models: Random Forest (RF), Gradient Boosting (GBM), and XGBoost. The classification task focuses on three real-world pavement types: smooth asphalt (with less than 10 years of service), degraded asphalt (with more than 15 years of service), and cement concrete pavement. Experimental results show that the proposed feature selection algorithm and classification models achieve high classification performance and fast execution speed. The system attains accuracy higher than 0.95 while reducing computational cost. These findings confirm the feasibility of deploying road-surface classification systems on low-cost devices for real-time pavement monitoring and highlight the importance of appropriate feature selection in balancing system accuracy and performance.]]>
