Dataset for Machine Learning-Based Link Adaptation in Highway NLOS Vehicular Networks

Machine learning–based link adaptation in vehicular networks remains limited by the lack of datasets that jointly capture PHY-layer dynamics and high-mobility effects under realistic propagation conditions. Most publicly available VANET datasets focus on traffic or application-layer metrics and do not provide packet-level channel descriptors required for adaptive modulation and coding research. This paper introduces VANET-LA Highway-NLOS, a physics-grounded dataset designed specifically for ML-driven link adaptation under obstruction-dominated highway scenarios compliant with IEEE 802.11p at 5.9 GHz. The dataset is generated using a MATLAB-based simulation framework that models Rayleigh fading with outage constraints, Doppler shifts up to 1500 Hz, delay spreads up to 1 µs relative to a 1.6 µs cyclic prefix, and Doppler-induced inter-carrier interference. A conservative, mobility-aware MCS selection strategy based on effective SNR, bandwidth scalability (2–10 MHz), and hysteresis stability ensures realistic adaptation behavior. The final dataset contains 234 imbalanced and 54 balanced samples described by nine synchronized features: bandwidth, SNR, Doppler shift, delay spread, MCS index, modulation order, coding rate, packet error rate, and achieved throughput. A structured class-equalization procedure eliminates the dominance of low-order MCS levels present in raw outputs, enabling unbiased supervised learning. Validation demonstrates physically coherent relationships between mobility, channel impairments, reliability, and throughput, confirming that mobility-induced distortions significantly constrain high-order modulation in highway NLOS conditions. By providing reproducible, PHY-consistent measurements aligned with realistic vehicular dynamics, VANET-LA Highway-NLOS offers a practical benchmark for developing and evaluating data-driven link adaptation algorithms aimed at improving reliability and spectral efficiency in high-speed V2X communications.