Prev
The IEEE ICDM TomTom Traffic Prediction for Intelligent GPS Navigation Contest is over. Contest was organized on TunedIT platform.
Over half a thousand researchers ventured to solve the problem of traffic in big agglomerations. The outcome? Accuracy improvement of up to 60% and an algorithm that can be used in GPS navigation devices in order to avoid traffic jams. And much more.
... Excessive traffic has become a development-related problem that affects everyone who lives in a city with a population of 50,000 or more, anywhere in the world. The complexity of the processes that shape the traffic flow is so high that only data mining algorithms, from the domains of structure mining, graph mining, data streams, large-scale and temporal data mining, may produce efficient solutions to these problems. The aim of the competition was to devise the best possible algorithms that would tackle the problems of traffic flow prediction, for the purpose of intelligent driver navigation and improved city planning. Over half a thousand participants tried to solve the problem and took part in the competition's three challenges:
1. Participants tried to predict traffic congestion basing the calculations on a series of measurements from 10 selected street network segments.
The goal was to make short-term predictions of future figures using historical ones for computations.
The author of the winning solution,
Alexander Groznetsky
from the Ukraine, designed an algorithm having an error rate of just 23% which is a result almost two times better than in case of the basic algorithm used in this task.
2. Modeling of the process of traffic jams forming during the morning traffic peak, when there are roadworks in progress, would be very useful for commuters.
Researchers submitted 1298 solutions, trying to predict the process basing on the data containing the coordinates of road network segments closed due to roadworks, accompanied by an indication of the segments where the first jams occurred.
The winning algorithm, created by a Warsaw University student
Łukasz Romaszko
, predicts the sequence of road segments where next jams will occur in the upcoming time. The winning algorithm describes the process more than 30% better than the existing one.
3. Reconstruct and predict traffic basing on real-time information from individual drivers.
Benjamin Hamner, a graduate of Duke University, was the most successful in this task.
Input data consisted of a stream of notifications from 1% of vehicles, sending in their current GPS locations in the city road network, sent every 10 seconds. The algorithm receives a stream and predicts traffic congestion on selected road segments for the next 30 minutes with an accuracy improved by 60% in comparison to the baseline. The method used for the creation of the algorithm and the solution itself creates a new opportunity for improvement in GPS navigation systems. The algorithm works in exactly the same way as those used in GPS car navigation system, such as TomTom Traffic. The algorithm proves very useful in selecting the optimal route to the destination.
Among nearly 5000 submitted solutions, very good results were achieved by the team from IBM T.J. Watson Research Center, USA which won the second place in the last two tasks.
The winners will be awarded 5000$ prizes and are all to have an opportunity to present their work at the IEEE International Conference on Data Mining in Sydney, Australia. The competition was sponsored by TomTom, the world's leading provider of portable GPS and car navigation systems, and held under the patronage of the Mayor of Warsaw. The competition is viewed as a valuable initiative which may help solve traffic problems not only in the Polish capital city, but also in other large agglomerations.
For more details visit: tunedit.org/challenge/IEEE-ICDM-2010
What TunedIT is
TunedIT (tunedit.org) is an international scientific platform for researchers and programmers in the fields of Data Mining, Computational Intelligence and Statistical Modeling. TunedIT runs periodically online contests - Challenges - that address selected real-world problems of data analysis. Competitions are open to the whole scientific community and attract participants from all over the world, from tens of different countries and hundreds of universities.