Interval prediction for short-term traffic forecasting using hybrid mode decomposition models

Traffic forecasting plays an important role for the implementation of intelligent transportation systems (ITS) and for ensuring appropriate traffic control schemes. Most existing forecasting algorithms provide point predictions without any indication to possibilities of variation in forecasts unless based on classical statistical methods. In this paper, we present a hybrid traffic prediction framework producing both point and interval forecasts utilising a combination of mode decomposition algorithms along with Artificial Neural Networks (ANN). A new decomposition algorithm known as Variational Mode Decomposition (VMD) is introduced to decompose traffic into a set of subseries which are modelled using ANNs. A quantile regression loss function is implemented to estimate prediction intervals. The performance of the proposed approach is evaluated using traffic flow data collected from a signalised junction in Dublin City (Ireland) and compared against a set of alternative hybrid models. Furthermore, the robustness of the proposed algorithm is established through consistency in performance when tested over varied prediction horizons.