Data mining competition being a part of IEEE International Conference on Data Mining 2010 (ICDM), Sydney, Australia, Dec 14-17. Sponsored by TomTom, the world's leading provider of portable GPS and car navigation systems. The task is to predict city traffic based on simulated historical measurements or real-time stream of notifications sent by individual drivers from their GPS navigators. Prizes worth $5,000 will be awarded to the winners.

Held under the patronage of the President of Warsaw, Mrs. Hanna Gronkiewicz-Waltz.

Introduction

Over the last century, number of cars engaged in vehicular traffic in cities has increased rapidly, causing many difficulties for all citizens: traffic jams, large and unpredictable communication delays, pollution etc. Excessive traffic became a civilization problem that affects everyone who lives in a city of 50,000 or larger, anywhere in the world. Complexity of processes that stand behind traffic flow is so large, that only data mining algorithms - from the domains of structure mining, graph mining, data streams, large-scale and temporal data mining - may bring efficient solutions for these problems. With the proposed competition, we want to ask researchers to devise the best possible algorithms that tackle problems of traffic flow prediction, for the purpose of intelligent driver navigation and improved city planning.

There are 3 tasks:

1. Traffic. Traffic congestion prediction, in an elementary setup of time series forecasting: a series of measurements from 10 selected road segments is given and the goal is to make short-term predictions of future values based on historical ones. This task is intended as an introductory one, simpler than the other two.
2. Jams. Modeling the process of traffic jams formation during morning peak in the presence of roadworks, based on initial information about jams broadcast by radio stations. Input data contain identifiers of road segments closed due to roadworks, accompanied by a sequence of segments where the first jams occurred. The algorithm should predict a sequence of segments where next jams will occur in the nearest future.
3. GPS. Traffic reconstruction and prediction based on real-time information from individual drivers. Input data consist of a stream of notifications from 1% of vehicles about their current GPS locations in the city road network, sent every 10 seconds. The algorithm receives this stream and predicts traffic congestion on selected road segments for the next 30 minutes. Large volumes of data are involved in this task, requiring the use of scalable data mining methods.

Tasks are independent. You may participate in all of them or in a chosen one.

You may submit solutions many times, for the whole duration of the challenge. They are evaluated instantly after submission and the result is published on Leaderboard, to allow comparison with other participants. However, to avoid overfitting to test data, there are two distinct evaluations: preliminary and final, that employ two separate subsets of test samples. During the contest, Leaderboard presents only preliminary results. Final results are published when the contest is closed and they may differ from preliminary ones. For each task, preliminary samples comprise 35% of test data and final samples - the remaining 65%.

If you have questions please ask them on the forum. News will be published in the News section, the most important ones will be posted also on Facebook and Twitter.

Traffic Simulation Framework

Datasets come from a highly realistic simulator of vehicular traffic, Traffic Simulation Framework (TSF), developed at the University of Warsaw. Simulations employ authentic map of the city of Warsaw, Poland, taken from OpenStreetMap project. TSF is based on the most well-known model of traffic flow by Nagel and Schreckenberg, which describes traffic on straight one-way road divided into some number of cells. Each cell may contain one car or may be empty. In each tick of a discrete clock, cars change their positions and speeds with carefully chosen rules. Nagel-Schreckenberg (NS) model was broadly examined. In particular, it was shown that this model can reproduce the phenomenon of traffic jams with unprecedented accuracy.

TSF extends the NS model to simulate traffic on an arbitrarily complex graph of road segments and incorporates new rules to model crossroads, traffic lights, multi-lane streets etc. Each car is an autonomous agent, with its own start and destination points, as well as preferences regarding its road behaviour. Road segments may differ in type (minor/major), default maximal velocity and number of lanes. It was confirmed by Warsaw citizens that TSF very accurately reproduces traffic jams, in the same places where they occur in reality.

After the contest, TSF will be released for public use, to enable post-challenge research.

Awards

Authors of the winning solutions will be awarded with prizes sponsored by TomTom, worth respectively:

• 1,000$ for track Traffic: 500$ of travel grant plus 500$ of cash prize
• 2,000$ for track Jams: 1,000$ of travel grant plus 1,000$ of cash prize
• 2,000$ for track GPS: 1,000$ of travel grant plus 1,000$ of cash prize

Travel grant covers travel and registration expenses related to participation in ICDM conference. If the first winner on a given track does not plan to participate in ICDM or his travel/registration costs will be lower, travel grant (or its part) will be awarded to the next best solution on this track.

Moreover, in 7 days after the end of the contest, participants may prepare short descriptions of their solutions and submit to contest organizers. Selected reports will be published in ICDM proceedings and presented at the conference during contest session. Reports should be formatted according to IEEE ICDM guidelines.

Schedule

• Jun 22, 2010: start of the challenge
• Sep 07, 2010: end of the challenge
• Sep 13, 2010: deadline for submission of short descriptions of solutions
• Dec 14-17, 2010: IEEE ICDM conference and contest session in Sydney

Organizing Committee

Marcin Wojnarski (Chair), TunedIT and University of Warsaw - marcin.wojnarski (at) tunedit.org
Paweł Gora, University of Warsaw
Hung Son Nguyen, PhD, University of Warsaw
Marcin Szczuka, PhD, University of Warsaw

If you have any questions please post them on the forum. We will reply as soon as possible.

Media patrons

Komputer Świat, Onet.pl, Students.pl, TVN Warszawa, Auto Świat, Przekrój, Niebezpiecznik.pl, Heise-online