Data Imputation for Connected Vehicle Speeds under Weather and Pavement Conditions

Reliable vehicle speed data are critical for traffic safety analysis, route planning, and transportation system management. Yet large-scale speed datasets are often incomplete, with substantial portions of data missing for extended periods. This study addresses missing data in real-world highway speed records collected from connected vehicles, where more than 50% of observations were absent for weeks to months between October and March. We evaluate a range of imputation methods, from classical regression models to state-of-the-art deep generative approaches that incorporate high-resolution weather and pavement condition data from a Road Weather Information System (RWIS). Our results show that while different models demonstrate strengths in different settings, Frequency-aware Generative Models for Multivariate Time Series Imputation (FGTI) consistently achieve the best overall performance.