Predicting Customer Attrition
## Attrited Customer Existing Customer
## CLIENTNUM 4.673173e-03 0.0001128593
## Customer_Age 5.379427e-03 0.0058420534
## Gender 6.989770e-03 0.0009719161
## Dependent_count 4.102233e-04 0.0006957406
## Education_Level -3.267433e-04 -0.0001757976
## Marital_Status 1.686215e-04 0.0009017120
## Income_Category 1.309759e-03 0.0003280176
## Card_Category -7.974519e-05 0.0003574737
## Months_on_book 4.658644e-03 0.0024841307
## Total_Relationship_Count 2.501008e-02 0.0296009187
## Months_Inactive_12_mon 2.502482e-02 0.0007412240
## Contacts_Count_12_mon 1.359501e-02 0.0034243404
## Credit_Limit 1.735621e-02 0.0133566315
## Total_Revolving_Bal 6.922971e-02 0.0440998282
## Avg_Open_To_Buy 1.648431e-02 0.0141157480
## Total_Amt_Chng_Q4_Q1 2.824554e-02 0.0065676617
## Total_Trans_Amt 1.350129e-01 0.0905490294
## Total_Trans_Ct 2.317611e-01 0.0887040048
## Total_Ct_Chng_Q4_Q1 7.381884e-02 0.0134476356
## Avg_Utilization_Ratio 5.209489e-03 0.0370197686
## MeanDecreaseAccuracy MeanDecreaseGini
## CLIENTNUM 0.0008446123 50.966525
## Customer_Age 0.0057660993 63.352681
## Gender 0.0019323792 16.218484
## Dependent_count 0.0006478947 24.486830
## Education_Level -0.0001991119 20.454767
## Marital_Status 0.0007834563 17.280836
## Income_Category 0.0004844410 21.301939
## Card_Category 0.0002868079 4.633761
## Months_on_book 0.0028264604 46.832402
## Total_Relationship_Count 0.0288743136 129.597981
## Months_Inactive_12_mon 0.0046362897 41.145641
## Contacts_Count_12_mon 0.0050526126 53.985864
## Credit_Limit 0.0139885484 63.326013
## Total_Revolving_Bal 0.0481085684 200.988496
## Avg_Open_To_Buy 0.0144867508 60.344190
## Total_Amt_Chng_Q4_Q1 0.0100470556 109.468020
## Total_Trans_Amt 0.0976511686 366.696662
## Total_Trans_Ct 0.1116344991 319.691758
## Total_Ct_Chng_Q4_Q1 0.0231385648 198.422527
## Avg_Utilization_Ratio 0.0318993894 114.874792
##
##
## Cell Contents
## |-------------------------|
## | N |
## | Chi-square contribution |
## | N / Row Total |
## | N / Col Total |
## | N / Table Total |
## |-------------------------|
##
##
## Total Observations in Table: 2990
##
##
## | val$Attrition_Flag_predicted
## val$Attrition_Flag | Attrited Customer | Existing Customer | Row Total |
## -------------------|-------------------|-------------------|-------------------|
## Attrited Customer | 391 | 91 | 482 |
## | 1548.957 | 252.436 | |
## | 0.811 | 0.189 | 0.161 |
## | 0.933 | 0.035 | |
## | 0.131 | 0.030 | |
## -------------------|-------------------|-------------------|-------------------|
## Existing Customer | 28 | 2480 | 2508 |
## | 297.686 | 48.514 | |
## | 0.011 | 0.989 | 0.839 |
## | 0.067 | 0.965 | |
## | 0.009 | 0.829 | |
## -------------------|-------------------|-------------------|-------------------|
## Column Total | 419 | 2571 | 2990 |
## | 0.140 | 0.860 | |
## -------------------|-------------------|-------------------|-------------------|
##
##
Table Prediction Error Descriptions
Error Type A: Predicted to be attrited, but are existing. Cost per
customer:$2,300 per year
Error Type B: Predicted to be existing, but are attrited. Cost per
customer:$2,300 per year
Table Prediction Errors
Error Type A: 91 x $2,300 = $209,300
Error Type B: 28 x $100 = $2,800
Conclusions
Benefit of model is that it detected 391 attrited customers that did
leave Assuming 30% agree to the incentive.
This means that 117 customers would be retained, with a mitigated
loss of $2,300 per employee.
This yields a total of $269,790.