Vehicle classification using accelerometer signals and machine-learning techniques

Abstract

Accurate vehicle classification is valuable for optimal traffic management, economic forecasting, emissions monitoring, and traffic enforcement (such as speed and weight limits). Existing approaches are often limited by expensive and intrusive sensors or camera-based systems, leading to privacy issues, installation complexity, and poor performance under adverse environmental conditions, such as rain, fog, and storms. This article proposes a cost-effective, non-intrusive classification method using pavement vibration captured through a single tri-axial accelerometer, intelligent feature extraction, and machine learning techniques. Class imbalance is mitigated using the adaptive synthetic sampling (ADASYN) technique. A stacked ensemble classifier combining Random Forest and XGBoost achieves 99.78% accuracy across eight vehicle classes. Model training over 50 independent iterations confirmed that the observed performance is robust and statistically significant. This approach demonstrates great potential for large-scale deployment in locations such as toll plazas, parking facilities, and logistics hubs, thus offering a low-cost solution that overcomes the limitations of conventional vehicle classification systems and adverse weather conditions. © 2025 Taylor & Francis Group, LLC.