Variables in GreatMart.xls Variable Definition Value CustNo Customer Number 00001 to 10000 Gender Gender of customer Male Female Age Age of customer 18 to 80 years old Months No of months that customer had been buying 6 to 54 months from GreatMart LastPorchAmt The last purchase amount made by the $10 to $260 customer TotalFurchAmt The total purchase amount made by the $35 to $1300 customer in the past 12 months Sales Whether the customer buy during any of Yes GreatMart sales event in the past 12 months No Complaint Whether the customer made any complaints to Yes GreatMart in the past 12 months No VIPMember Whether the customer is a VIP member in the Yes past 12 months. (A VIP member status is given |No based on total purchase amount in the past 3 years as well as frequency of purchases. A VIP member will receive more discount coupons.) Dormant Whether the customer was dormant in the past Yes 6 months (i.e., the customer had not made any No [purchase in the past 6 months) A total of 2144 (or 21.44%) out of 10000 customers were dormant in the past 6 months. For this analysis, three decision trees were constructed using CHAID, CART and C5.0. The settings of the data partition and decision trees used are as follows: Data Partition Training the model: 70% Testing the model: 30% No changes to the other settingsCHAID and CART Minimum records in parent branch (%): 10% Minimum records in child branch (%): 3% No changes to the other settings C5.0 Minimum records per child branch: 200 No changes to the other settingsAnalysis Annotations Collapse A * Expand All iR-Domat Reaulta for output held Dormant SRI-Derman 3 Individual Models Comparing SR-Dormant with Dormant "Partition' 1_Training _Testing Correct 6.117 87 76% 2478 86.35% Wrong 853 12 24% 352 11.62% Total 3,030 Coincidence Matrix for ER-Dormant (rows show actuals) CART (R) Partition' = 1_Training No Yes 4,228 230 CART 623 "Partition' = 2_Testing No Yes 2,308 Yes - Comparing SR1-Dormant with Donmart C5.0 (C) "Partition' 1_Training 2_Testing Correct 6004 26 14% 2.601 85.84% Wrong 856 13 85% 429 14. 15% Total 6,970 3,030 Co nodence Matin for $R1-Dormant (rows show actuals "Partition' - 1_Training No Yes No 5,177 201 CHAID Tes 645 827 "Paration' - 2_Testing No Yes CHADD (RI) 2 274 124 Yes 305 327 Comparing SC-Dormant with Dormant Partition' 1_Training _Testing Correct 6.241 8954%% 2,717 89.67% Wrong 729 10 46%% 313 10.33% Total 6.970 3030 Coincidence Matrix for BC-Dormant (rows show actuals) "Partition' = 1_Training No Yes 3,332 128 Partition C5.0 Yes 603 Dormant = "Yes" "Partition' =_Testing No Yes 2345 53 Yes 260 372 10 Decide Testing Figure 3: Analysis Node Output and Lift Charts ("Yes" represents dormant customers and "No" represents non-dormant customers)Dormant Node O Categor NO 78.560 7856 21.440 2144 Total (00.000 10000) Months Improvement=0.089