DP-LRT: An Urban Short-term Traffic Speed Forecasting Method Based on Data Driven

Abstract

Until recently, most short-term traffic speed forecasting algorithms have been applied on freeway, arterial or corridor. Short-term traffic speed forecasting on urban road network formed a more complex problem than freeway predictions due to constraints such as signalization. This paper proposes a Date Pattern and Likelihood Ratio Test statistics based (DP-LRT) method for urban short-term traffic speed forecasting from large scale taxi GPS data. Experiment showed that the proposed method reduced the mean absolute percent error and improved the prediction success rate compared with K-NN method, which was a representative data-driven method.