Travel Time Prediction of Taxis using Tensor Factorization

Abstract

The analysis of data streams offers a great opportunity for development of new methodologies and applications in area of Intelligent Transportation System. Travel time prediction of a path is an important task in route planning and navigation applications. As more GPS floating car data has been collected to monitor urban traffic, GPS trajectories of floating cars have been frequently used to predict path travel time. Solving this problem involves two major impediments which are the facts that firstly many road segments may not be traveled by any GPS equipped vehicles in given time slot and in most cases a trajectory exactly traversing a query path cannot be found either. Second for the fragment of a path with trajectories there are multiple ways of combining the trajectories to estimate corresponding travel time. To address these challenges different drivers travel times on different road segments in different time slots is modeled with a three dimension tensor. Combined with geospatial, temporal and historical contexts learned from trajectories the tensor's missing values are filled through a context-based tensor decomposition approach. Therefore this paper aims to solve the problem that given a travel time prediction query, the congestion patterns around the query path are identified from historical trajectories, then its travel time is inferred in near future. Experiment results using Uber GPS traces[1] of black cab pickups which comprises of 14500 trips (after data cleaning) in areas of San Francisco shows that the prediction has on average 3.36 minutes of error on trips of duration 6 to 25 minutes.