Challenges / IEEE ICDM Contest: TomTom Traffic Prediction for Intelligent GPS Navigation

Status Closed
Type Scientific
Start 2010-06-22 18:00:00 CET
End 2010-09-07 23:59:59 CET
Prize 5,000$

Registration is required.

Summary

The data mining challenge, IEEE ICDM Contest: TomTom Traffic Prediction for Intelligent GPS Navigation, is over now. It attracted 575 participating teams, of whom over 100 submitted solutions, most of them many times: the total number of solutions was close to 5000. Best algorithms achieved nearly 3-fold improvement over baseline solutions in predicting traffic congestion and jams. We thank all participants for their effort and congratulate the winners! We also thank TomTom company and the President of Warsaw for their support of the contest.

The Winners

Traffic:

  1. Alexander Groznetsky, Ukraine (alegro)
  2. Carlos J. Gil Bellosta, Datanalytics, Spain (datanalytics)
  3. Benjamin Hamner, Duke University, USA (hamner)

Jams:

  1. Łukasz Romaszko, University of Warsaw, Poland (lukasz21212121)
  2. Jingrui He, Qing He, Grzegorz Swirszcz, Yiannis Kamarianakis, Rick Lawrence, Wei Shen and Laura Wynter, IBM T.J. Watson Research Center, USA (trafficlab)
  3. Kenneth Shirley, Carlos Scheidegger, Ji Meng Loh and Suhris Balakrishnan, AT&T Labs Research, USA (regress)

GPS:

  1. Benjamin Hamner, Duke University, USA (hamner)
  2. Wei Shen, Yiannis Kamarianakis, Jingrui He, Qing He, Rick Lawrence, Grzegorz Swirszcz and Laura Wynter, IBM T.J. Watson Research Center, USA (trafficlab)
  3. Andrzej Janusz, University of Warsaw, Poland (NosferatoCorp)
Winners of ICDM 2010 Contest in Sydney

After ICDM Contest Workshop in Sydney. From left: Paweł Gora, Marcin Wojnarski (organizers), Łukasz Romaszko, Benjamin Hamner, Carlos J. Gil Bellosta (winners), Ralf-Peter Schäfer (TomTom)

Top Solutions

Top solutions are now described on TunedIT Blog, see Part 1 and Part 2 of the blog post. Comment below the posts if you want to ask the authors any questions.

ICDM Contest Workshop

During ICDM conference, the ICDM Contest Workshop will be organized, where top solutions will be presented by their authors. The Workshop will take place on December 14th (Tue), at the ICDM conference venue at the University of Technology, Sydney, Building 5 (Haymarket Campus). Key points presented at the Workshop will be repeated on the following day (Wed, Dec 15th, 14:00 - 15:30) during shorter ICDM Contest Overview Session, as a part of the main conference programme.

We invite everyone to attend!

Detailed Programme

13:30 - 14:00    Introductory Report
Marcin Wojnarski, Paweł Gora, Marcin Szczuka, Hung Son Nguyen, Joanna Swietlicka, Demetris Zeinalipour

14:00 - 14:30    TomTom: Insights about TomTom's Traffic Technology
Ralf-Peter Schäfer

14:30 - 15:00    Task 1 (solution 2): A convex combination of models for predicting road traffic
Carlos J. Gil Bellosta

15:00 - 15:30    Task 1 (solution 3): Predicting Future Traffic Congestion from Automated Traffic Recorder Readings with an Ensemble of Random Forests
Benjamin Hamner

15:30 - 16:00    Coffee Break (30 min)

16:00 - 16:30    Task 2 (solution 1): IEEE ICDM 2010 Contest: Traffic Prediction - Jams
Łukasz Romaszko

16:30 - 17:00    Task 2 (solution 2): Ensemble-based Method for Task 2: Predicting Traffic Jams
Jingrui He, Qing He, Grzegorz Swirszcz, Yiannis Kamarianakis, Rick Lawrence, Wei Shen, Laura Wynter

17:00 - 17:30    Task 3 (solution 1): Predicting Travel Times with Context-Dependent Random Forests by Modeling Local and Aggregate Traffic Flow
Benjamin Hamner

17:30 - 18:00    Task 3 (solution 2): Traffic Velocity Prediction Using GPS Data: IEEE ICDM Contest Task 3 Report
Wei Shen, Yiannis Kamarianakis, Laura Wynter, Jingrui He, Qing He, Rick Lawrence, Grzegorz Swirszcz

Post-challenge Research

All challenge resources are published now in the challenge folder in Repository and everyone can use them as a basis for new research. They are divided into three subfolders: Traffic, Jams and GPS.

The following resources are available:

  • Traffic Simulation Framework - the simulator used to generate the input data, along with its instruction
  • Street graph used by the simulator
  • Evaluation procedures for preliminary and final tests
  • Training data
  • Test data: split into public part (inputs) and target decisions
  • Example solutions
  • Final solutions of all participants

Post-challenge Submissions

Post-challenge submissions per se are not supported. Instead, TunedIT provides a versatile framework for evaluation of new solutions against any evaluation procedure and dataset, not only the ones used for the challenge. Thus, you may prepare derived versions of challenge resources and compare test results with the original ones. You may also control access rights to your resources and remove once uploaded solutions. To use the framework, upload a solution file to your home folder in Repository and evaluate it with TunedTester, giving the following names of the evaluation procedure and datasets in TT window (plus give the full name of your algorithm, as shown on its Repository page):

Traffic:

  • Evaluation: ICDM/2010/traffic/eval1prelim.jar:traffic.Eval1Prelim or ICDM/2010/traffic/eval1final.jar:traffic.Eval1Final
  • Dataset: ICDM/2010/traffic/test_priv.txt

Jams:

  • Evaluation: ICDM/2010/jams/eval2prelim.jar:traffic.Eval2Prelim or ICDM/2010/jams/eval2final.jar:traffic.Eval2Final
  • Dataset: ICDM/2010/jams/test_priv.txt

GPS:

  • Evaluation: ICDM/2010/gps/eval3prelim.jar:traffic.Eval3Prelim or ICDM/2010/gps/eval3final.jar:traffic.Eval3Final
  • Dataset: ICDM/2010/gps/3test_priv.txt

If you wish, you may choose to send results to Knowledge Base. They can be viewed and analyzed afterwards by everyone on KB pages. For example, the following links generate post-challenge leaderboards with results corresonding to the same evaluation setup as used in the challenge (the same evaluation procedures and datasets):

When you open KB page, notice that there may be several evaluation procedures or datasets on the drop-down lists. Every choice generates a different chart. Also, there is the Raw Results tab where all the results are presented in tabular form and can be downloaded as CSV files. Raw Results tab contains also hyper-links to Repository pages of particular algorithms and datasets.

Public files

Copyright © 2008-2013 by TunedIT
Design by luksite