White-Box Modeling of V2X Link Performance Using Stabilized Symbolic Regression

TL;DR

This paper introduces a stabilized symbolic regression method to create interpretable, compact models for vehicle-to-everything link performance, achieving high accuracy and computational efficiency suitable for real-time applications.

Contribution

The work presents a novel stabilized symbolic regression framework that accurately models V2X block error rate with interpretable expressions, outperforming fixed-form regressions and matching neural network accuracy.

Findings

Achieved R^2 = 0.8684 and MSE = 2.08e-2 on test data.

Final symbolic expression has only 158 nodes, enabling fast inference.

Outperforms traditional regressions and rivals neural networks in accuracy.

Abstract

Reliable modeling of block error rate in vehicle-to-everything wireless networks is critical for designing robust communication systems under dynamic mobility and diverse channel conditions. Traditional machine learning approaches, such as deep neural networks, achieve high predictive accuracy but lack interpretability and impose significant computational costs, limiting their applicability in real-time, resource-constrained environments. In this work, we propose a stabilized symbolic regression framework to derive compact, analytically interpretable expressions for block error rate prediction. Trained on realistic vehicle-to-everything simulation data, the symbolic regression framework for vehicle-to-everything model accurately captures nonlinear dependencies on key system parameters, including signal-to-noise ratio, relative velocity, modulation and coding schemes, number of demodulation reference signal symbols, and environmental factors (line of sight/non-line of sight). Our final symbolic expression comprises only 158 nodes, enabling ultra-fast inference suitable for embedded deployment. On the test set, the symbolic regression framework for vehicle-to-everything model achieves a coefficient of determination $R^2 = 0.8684$ and mean squared error $= 2.08 \times 10^{-2}$ in the original block error rate domain, outperforming conventional fixed-form regressions and offering comparable accuracy to neural networks while remaining fully interpretable. Overall, the proposed Stabilized Symbolic Regression Framework for V2X combines predictive performance, physical fidelity, and computational efficiency thus providing a powerful tool for real-time V2X communication system design, adaptive resource allocation, and rapid scenario evaluation.