Neuro-Symbolic Hybrid Ensemble for Spatio-Temporal Traffic State Inference and Priority-Aware Decisioning in VANETs

Abstract

Modern vehicular networks continuously generate large volumes of real-time data, including vehicle speed, GPS coordinates, signal strength, packet delay, congestion levels, and available bandwidth, all of which are essential for efficient traffic management and network stability. Conventional traffic monitoring approaches, which rely on manual analysis or fixed threshold mechanisms, often result in delayed responses, limited prediction accuracy, and suboptimal resource utilization. To address these limitations, this study proposes a robust machine learning framework based on Classification and Regression Trees (CART) for accurate prediction of vehicle priority levels and traffic density. While existing hybrid models such as SVM-CART and RF-CART offer baseline performance, they often struggle to effectively capture complex non-linear relationships and may suffer from overfitting in diverse traffic scenarios. To overcome these issues, a Hybrid Stacked Ensemble CART (HSE-CART) model is introduced, integrating Random Forest (RF) and Hierarchical Soft Tree (HSTree) as base learners, with Logistic Regression (LR) and Linear Regression (LinR) as meta-learners for classification and regression tasks. The proposed approach enhances predictive capability by effectively balancing bias and variance while modeling intricate feature interactions. Experimental results demonstrate significant improvements across performance metrics, including accuracy, precision, recall, F1-score, and R², highlighting its effectiveness in enabling reliable, real-time traffic prediction and intelligent vehicular network management.