Features

AusDM Analytic Challenge:

The Australasian Data Mining Conference is devoted to the art and science of intelligent data mining: the meaningful analysis of (usually large) data sets to discover relationships and present the data in novel ways that are compact, comprehendible and useful for researchers and practitioners.

This conference will bring together from the Data Mining and Business Analytics community researchers and practitioners to share and learn of research and progress in the local context and new breakthroughs in data mining algorithms and their applications.

Data mining competitions provide a great arena for knowledge transfer, benchmarking and idea generation, with competitors generally more than happy to share their methods and techniques for all to try. They are also often a valuable source of real world data where theory can be put to the test and methods tested on different domain data.

This year sees the staging of the first AusDM Analytic Challenge. The theme of the challenge is 'ensembling' which is beginning to be seen as a necessity if you want to stretch the accuracy of predictive models.

When it comes to predicting human behaviour, it is clear that no single algorithm dominates - which is not surprising as human behaviour is a complex thing to model. What is evident though is that different algorithms that seemingly give a similar performance will complement each other when combined, resulting in more accurate overall predictions.

In the last three years or so, these 'ensembling' methods have become a popular tool in predictive competitions, but how the individual models are combined is often somewhat arbitrary. The AusDM Analytic Challenge is to address this question - what is a good way to combine models - and there is a $1,000 prize to the winner.

The added interest is that the data being used comes from the two top teams in the Netflix Prize, so hopefully it will also answer the question many people want to know, just how low could the error have gone if these two teams had joined forces.

The challenge has already started and ends on November 22nd. The winners will be anounced at the conference.

For more details and to enter the challenge, please visit

www.tiberius.biz/ausdm09

Phil Brierley
AusDM Competition Chair