Skip to main content

Table 7 Basic statistical analysis of features used by the RPI-CP algorithm

From: On information propagation in mobile call networks

Feature Number Churner, DS-1 Non-churner, DS-1 NP-err, DS-1 Churner, DS-2 Non-churner, DS-2 NP-err, DS-2
1 1.72 ± 1.46 1.9 ± 1.9 0.48 14 ± 13.8 14 ± 14 0.48
2 0.27 ± 0.59 0.32 ± 0.84 0.48 2 ± 2.6 1.9 ± 2.5 0.48
3 0.3 ± 0.7 0.37 ± 0.99 0.48 2.74 ± 4.14 2.4 ± 3.7 0.48
4 1.32 ± 2.76 2.02 ± 7.74 0.48 19.5 ± 35.2 18.3 ± 32.5 0.48
5 0.02 ± 0.15 0.02 ± 0.16 0.5 0.16 ± 0.51 0.19 ± 0.63 0.5
6 0.003 ± 0.06 0.006 ± 0.08 0.5 0.02 ± 0.16 0.02 ± 0.15 0.5
7 0 ± 0.02 0 ± 0 0.5 0.01 ± 0.11 0.011 ± 0.1 0.5
8 0.03 ± 0.17 0.03 ± 0.18 0.5 0.19 ± 0.59 0.22 ± 0.71 0.5
9 44 ± 193 84 ± 306 0.44 2,589 ± 5,875 3,143 ± 7,925 0.44
10 69 ± 301 132 ± 486 0.43 3,484 ± 7,872 4,225 ± 10,620 0.43
11 6.3 ± 86.4 18 ± 155 0.49 583 ± 2,326 741 ± 2,914 0.49
12 0.002 ± 0.05 0.001 ± 0.03 0.5 0.002 ± 0.04 0.002 ± 0.08 0.5
13 0.002 ± 0.05 0.001 ± 0.03 0.5 0.001 ± 0.04 0.001 ± 0.05 0.5
14 0.64 ± 1.05 0.65 ± 1.46 0.48 0.95 ± 0.82 0.9 ± 0.8 0.48
15 0.02 ± 0.12 0.02 ± 0.12 0.5 0.01 ± 0.06 0.011 ± 0.05 0.5
16 29 ± 138 50 ± 197 0.45 140 ± 316 155 ± 342 0.45
17 45 ± 214 76 ± 299 0.45 188 ± 424 208 ± 457 0.45
18 6 ± 86 17 ± 149 0.49 405 ± 1,464 465 ± 1,577 0.49
19 0.15 ± 0.32 0.13 ± 0.29 0.48 0.14 ± 0.17 0.13 ± 0.17 0.48
20 0.15 ± 0.32 0.13 ± 0.29 0.48 0.16 ± 0.18 0.15 ± 0.18 0.48
21 0.001 ± 0.03 0.001 ± 0.02 0.5 0 ± 0 0 ± 0.01 0.5
22 0.001 ± 0.03 0.001 ± 0.02 0.5 0 ± 0.01 0 ± 0.01 0.5
23 720 ± 1,187 848 ± 1,817 0.37 6,260 ± 7,797 7,511 ± 9,909 0.37
24 514 ± 922 894 ± 2,875 0.39 5,603 ± 8,495 7,150 ± 16,682 0.39
25 1,234 ± 1,678 1,742 ± 3,571 0.33 11,863 ± 14,121 14,661 ± 22,028 0.33
26 7.1 ± 8.3 7.23 ± 8.51 0.47 51 ± 51 57 ± 59 0.47
27 5.7 ± 8.6 7.58 ± 17.66 0.47 47 ± 57 55 ± 113 0.47
28 12.8 ± 13.9 14.8 ± 21.43 0.45 97 ± 100 113 ± 147 0.45
  1. For each feature and for each data set, we list the following three values. First, the mean value of the considered feature for churners, i.e., the users that churned during the testing period. The confidence interval of this mean value is given by one standard deviation. Second, the mean value of the considered feature for non-churners, i.e., the users that did not churn during the testing period. Again, the associated confidence interval is given by one standard deviation. Third, we provide the value of Neyman–Pearson error (denoted by NP-err)