Wednesday, September 30, 2009

The results have been posted! Congrats to the winners!

Thank you all for your hard work! Congratulations to the winner and the runners up.

Hope to see some of you at INFORMS. We will have a recognition session on Saturday 4-5PM during the Data Mining Workshop and presentations of the work on Tuesday.
I am currently trying to finalize the talks.



  1. It appears to me that the top 4 results in both categories differ only by chance. The fifth is "iffy." My guess is that it really is slightly inferior, but it could just be unlucky, I suppose.

    So who's the best overall? I suppose I'll promote using average score to determine this, while others will go for average rank. There's a way of spinning it to please everybody.

    Did anybody make non-trivial use of the historical patient data? I didn't.

    Thanks, Claudia.

  2. I used current visit and the preceding one. I created some variables by averaging all previous info, but seems did not contribute much.

    I found the length of stay and the costs of current visit (like ANCILCHARGES) are good predictors. While in real case they are not available. Therefore, I think our results overestimated the performance. I saw some publications on predicting mortality in patients with heart failure. The highest AUC is around 0.85.

    By the way, it seems that not much people look at this site since the comments are hidden under the page.