Modeling Traffic Flows Time Series Using Simplified D-Vines

Accurate forecasting of traffic flows is essential for trans-portation planning. However the modeling of these data is challeng-ing due to the presence of strong seasonalities and complex temporal dependence. This study proposes a vine-transform copula time series model applied to hourly traffic counts derived from mobile phone Origin– Destination data in the province of Brescia, Italy. The analyzed approach separates marginal dynamics by modeling counts using a negative bino-mial regression with seasonal covariates and serial dependence by captur-ing it through a D-vine copula applied to probability integral transforms. The first innovation regards vine copula time series models to traffic count data using negative binomial margins, proposing a simplified D-vine structure that can capture hourly, daily, and weekly dependence. Second, we show the effectiveness of this approach applied to large-scale mobility data. The model achieves a RankGraduationAccuracymea-sureequalto0.9689,andthecoverageindicatesthatabout88.79%oftheactualobservedvaluesfallwithin90%predictionintervals.