Travel Time Estimation Using Floating Car Data

This project explores the use of machine learning techniques to accurately predict travel times in city streets and highways using floating car data (location information of user vehicles on a road network). The aim of this report is twofold, first we present a general architecture of solving this problem, then present and evaluate few techniques on real floating car data gathered over a month on a 5 Km highway in New Delhi. 1 Floating Car Data Based Traffic Estimation Data used to estimate the travel times on road networks come in two varieties, one being fixed sensors on the side of the road such as magnetometer detectors or highway cameras [5, 6]. The second method is floating car data (FCD). Floating car data are position fixes of vehicles traversing city streets throughout the day. The most common type of FCD comes from taxi’s or delivery vehicles which are on main arterial roads and highways throughout most of the day. This second approach has many positive and negative attributes that must be dealth with to provide accurate travel time inference. First, FCD is perhaps the most inexpensive data to attain, since many taxi services, and delivery companies automatically gather this data on their vehicles for logistic purposes. Second, position fixes are generally very accurate, since GPS is used and this has high accuracy. There are however many disadvantages as well. First, FCD is usually sampled infrequenty, on the order of 2-3 minutes. The reason for this is that taxi or delivery companies do not need such fine time granularity of their vehicles position. Therefore quiet a bit of preprocessing needs to take place in order to “snap ” sets of points onto the proper streets with the possibility that multiple paths might have lead to the same pair of traverse points. Another disadvantage of this method is that a high density of data is required to get meaningfull travel time predictions for a given road network. Keeping all this in mind, constructing more and more accurate travel time predictions can be a fruitful Algorithms/Machine Learning/Statistical Modelling problem with various problems to tackle.