The first place and EUR 1.200,– went to Team A. Merćep, R. Vaser and S. Begusic, Computer Science students from Zagreb. For the problem at hand, they not only used a single model, but combined a large number of predictive models trained with powerful graphical processing units (GPUs). Since real data always contain missing entries and unknown values, we manually defined some additional rules that reflect the properties of the process for generating exhaust gases in vehicles.
Second (EUR 800,–) and third (EUR 500,–) place went to M. Kampelmühler and S. Müller from Graz, who achieved only slightly less impressive results with their statistical and machine learning approaches.
This year, 50 people took part in our Data Science Challenge, which represents an increase of 25% over the previous year. The participants used a variety of different methods to predict NOx emissions from vehicles and combustion engines. Similar to 2017, most participants reside in Austria. However, as a third of our participants come from abroad, we see a growing international community interested in our small competition.
The data set of the competition and the evaluations of the results are now available for public download. If you are able to exceed the benchmark of a mean absolute error of 36.85 for the validation dataset, we would be pleased if you send an e-mail to firstname.lastname@example.org with your result. For exceptional results, we provide you with a goodie!
We are looking forward to our next Data Science Challenge, which hopefully will include more of everything: more rewards, more participants and even more complex problems to solve!