Java Data Mining Concepts - Java Data Mining: Strategy, Standard, and Practice - page 149

Table 7-5 Lift computations table

Number of
customers likely
to attrite

Cumulative number
of customers likely
to attrite

Quantile
number

Cumulative
quantile lift

Cumulative gain

1 70 70 70/19 3.684 70/190 36.8%

2 40 110 110/38 3.289 110/190 57.9%

3 25 135 135/57 2.368 135/190 71.5%

4 15 150 150/76 1.974 150/190 79.4%

5 12 162 162/95 1.705 162/190 85.7%

6 8 170 170/114 1.491 170/190 89.8%

7 7 177 177/133 1.331 177/190 93.4%

8 5 183 183/152 1.204 183/190 96.5%

9 5 188 188/171 1.099 188/190 99.0%

10 3 190 190/190 1.000 190/190 100%

Using the classification model, the first quantile contains the top
100 customers that are predicted to be attriters. Comparing the pre-
diction against the known actual values, we find that the algorithm
was correct for 70 of these 100 customers. Therefore, the lift for the
first quantile is 70/19 3.684, where 70 is the number of attriters

found using the classification model and 19 is the number of customers
that would have been found given a random selection of customers.
Similarly, the cumulative gain value for this first quantile is the
percentage of the attriters in this quantile, that is, 70/190 0.368. In

Table 7-5, observe that the cumulative quantile lift values gradually
decrease, because the addition of each quantile includes fewer probable
cases, and the last quantile has lift value 1 because it includes all
1,000 cases. Cumulative gain values gradually increase, because the
addition of each quantile increases the proportion of attriters, and the
last quantile has cumulative gain of 100 percent because it includes all
1,000 cases.
In this example, suppose that ABCBank wants to launch a cus-
tomer retention campaign with a limited budget that can retain at
least 50 percent of the attriters. Here, the user can select the 200
customers in the first two quantiles whose cumulative gain is 57.9
percent and has a lift of 3.289.