2009 UC San Diego Data Mining Contest

The History of the UCSD Student Data Mining Contest

Since 2004, UCSD and FICO have hosted this data mining contest with the goal of providing students an opportunity to test out their data mining skills.

2009 Contest

In the sixth UCSD/FICO datamining contest, international teams made their strongest showing yet, taking all but one of the twelve prizes. We had competitors from all inhabited continents except Africa, comprising a total of 301 teams, with 151 international teams. In both tasks, contestants were asked to predict the presence of an anomaly in e-commerce transaction data. The data in the "hard" task lacked some of the structure present in the "easy" task's data.

The winners were:

2008 Contest

In its fifth year, the contest attacted international attention, with competition from Europe, Asia, and North America. As in the past, the UC schools made a strong showing. Contestants were asked to classify data from a scientific experiment. The first task was to maximize accuracy given a fully labeled training data set. The second task was to maximize the F1-score given a partially labeled training set.

The winners were:

2007 Contest

In its forth year the contest continued to attract competition from across the US, with schools ranging from Oregon State University, Stanford, and Harvey Mudd College to the University of Cincinnati, the University of Central Florida, SUNY - Univ. at Buffalo, and the Medical University of South Carolina. As always, the UC campuses had a strong presence. The contest was based on San Diego housing data, provided by DataQuick. Contestants were asked to predict whether or not a particular property would refinance, and how much it would sell for.

The winners were:

2006 Contest

In the third year, we went national with the competition. We had 58 teams compete from a number of American universities: all of the UC campuses, U Penn, Oregon State, U Texas, U Michigan, Rutgers, Suny Binghamton, Central Florida, OSU (Oklahoma and Oregon), Stanford, MIT and Georgia State.

The contest focused on text processing: document classification and word prediction. The classification task was hotly contested with a number of teams finishing very close to the top score. Two undergraduate teams (Bussewaffenlos and decaff) finished second and third overall beating out all but one graduate team (classified). UC Los Angeles dominated the Word Prediction task finishing with three of the top four teams overall (ShowMeThePrizes, GoldSeekers, and MoneyMoneyMoney). Team decaff (Chuan Sheng Foo) finished an impressive third to take the top undergraduate award.

The winners were:

2005 Contest

In the second year, we opened the contest up to all of the University of California campuses (Berkeley, Davis, Irvine, Merced, Riverside, San Francisco, San Diego, Santa Barbara, & Santa Cruz). The data set was a set of time series where each time series could be thought of as an account profile. The contestants had to predict if and when the account changed from a 'good' state to a 'bad' state. We had over 90 teams representing 8 UC campuses participate!

The winners were:

2004 Contest

In the our first year, a pilot contest was created for the UC San Diego community. The contest data set was mass spectrometry measurements from the emissions of cars and trucks. The first task was to predict if the emissons came from a car or a truck. The second task was to predict the negatively-charged mass spectrometry measurements given the positively-charged measurements. We had 10 undergraduate and 20 graduate teams compete.

The winners of the pilot competition were:

Brought to you by UCSD and FICO Web Site Design: Eric Christiansen, Eric Wiewiora, Douglas Turnbull and Dustin Boswell (adapted from Steve Gunn)