Lightweight Machine Learning Framework for Latency Prediction in Automotive Communication Networks

The increasing complexity of automotive systems—driven by autonomous driving, V2X communications, and advanced infotainment—demands low-latency, high-bandwidth in-vehicle communication networks. This paper presents a machine learning-based framework for predicting physical-layer latency across three widely used automotive communication protocols: CAN-FD, FlexRay, and 1000BASE-T1. The model is trained on a dataset generated via LTspice simulations, which incorporate transmission line theory to represent electrical parameters such as resistance, inductance, and capacitance. Among various models evaluated, Ridge Regression achieved the best performance, improving R2 scores by approximately 31% for CAN-FD, 30.8% for FlexRay, and 30.0% for 1000BASE-T1. The proposed approach reduces input dimensionality from 12 to 6 features through systematic feature engineering, offering a scalable and accurate surrogate model. This framework significantly reduces simulation time and design complexity, providing a valuable tool for early-stage latency estimation in automotive networks. The findings contribute toward building reliable, real-time communication systems for next-generation connected and autonomous vehicles. This framework reduces design cost by minimizing the need for iterative full-wave simulations.