A Hybrid PSO–Random Forest Framework for Real-Time Vehicle Classification Using Roadside LiDAR

Intelligent transportation systems (ITSs) are increasingly reliant on precise and efficient vehicle classification to support traffic management, safety applications, and infrastructure planning. However, existing classification models often suffer from inefficiencies in feature selection, limited scalability, and suboptimal performance in real-time environments. To address these challenges, this article proposes a hybrid particle swarm optimization (PSO)-enhanced random forest (RF) framework for real-time vehicle classification using roadside light detection and ranging (LiDAR) data. A comprehensive vehicle classification standard is formulated by synthesizing national guidelines and the Highway Capacity Manual, and a curated dataset derived from the SDUITC database, comprising 10 002 manually labeled samples, is employed for training and validation. The PSO algorithm is used to optimize key hyperparameters of the RF classifier, enhancing the model’s predictive performance by reducing redundancy and dimensionality. Experimental results demonstrate that the proposed PSO–RF model achieves a remarkable overall accuracy of 91.43% and a macro- ${F}1$ -score of 0.9124 while maintaining a classification time of 6 ms/frame. Comparative analysis with baseline models, including support vector machine (SVM), convolutional neural network (CNN), standalone RF models, and genetic algorithm-optimized RF (GA-RF) model, confirms the superiority of the proposed method in terms of both accuracy and real-time performance. These findings indicate that the PSO–RF framework is well-suited for deployment in cooperative vehicle–infrastructure systems, offering a robust, efficient, and scalable solution for real-time vehicle classification in smart transportation networks.