http://www.data-mining-cup.com/2006/Wettbewerb/Aufgabe/1190662447/

Problem:

• Predict which items obtain higher price than average based on listing attributes

Methodology:

1. 1. A few covariate additions/modifications - 4077 keywords from listing & listing subtitles were converted to binary variables with a script
2. 2. Parameters were tuned only within training sample
3. 3. Final model: 1000 random trees on 8000-sample training data, ap=0.01
4. 4. Total modeling time less than one day

Results:

• Final score (right - wrong): 5028
• Contest winner score: 5020
• We would have ranked 1st out of 579 participants & 191 entries

Data Mining Cup 2006 Results