ppt课件-lessonlearntfromtheucsddataminingcontest.pptVIP

Lesson learnt from the UCSD datamining contest Richard Sia2008/10/10 Background information Since 2004 2004 – UCSD 2005 – UC campus (loan account good/bad status) 2006 – US schools (text classification problem) 2007 – US schools (property refinance prediction) 2008 – worldwide (physical science experiment data) Sponsored by FairIssac (prize money) Tracks Undergraduate / postgraduate Classic classification task / Data-specific task Training / Quiz / Test data Dataset: Physical Science Experiment Dataset Class label (1, -1) 20 real values features Task 1: Standard classification Training: 40,00