http://www.data-mining-cup.com/2007/Wettbewerb/Aufgabe/1190662447/

Problem:

• Maximize profit, predicting which coupons customers will use
• Merchant loses $2 per unredeemed coupon, gains $3/coupon A, $6/coupon B

Methodology:

1. 1. Optimal thresholds for profitability found using training sample
2. 2. One model built for each coupon type
3. 3. 400 random trees on 50,000-sample training data, ap=0.05
4. 4. Thresholds were optimized to maximize campaign ROI
5. 5. Total modeling time less than 1 day

Results:

• Our final score $8013
• Contest winner score: $7907
• We would have ranked 1st out of 688 participants & 246 entries

Data Mining Cup 2006 Results