One-Step Dynamic Classifier Ensemble Model for Customer Value Segmentation with Missing Values
Table 3
Comparison of performance in “German” dataset with MCAR type MVs.
Missing level
Evaluation criteria
LMF
ODCEM
KI-SVM
MS-SVM
EM-SVM
RI-SVM
%
Total accuracy
0.6767 (3)
0.6987 (1)
0.6700 (4)
0.6933 (2)
0.6600 (5)
0.6533 (6)
AUC
0.8402 (4)
0.8592 (1)
0.8274 (6)
0.8583 (2)
0.8456 (3)
0.8334 (5)
Type I accuracy
0.5513 (4)
0.6124 (1)
0.5239 (6)
0.6069 (2)
0.5529 (3)
0.5298 (5)
Type II accuracy
0.7304 (3.5)
0.7357 (2)
0.7412 (1)
0.7304 (3.5)
0.7059 (6)
0.7108 (5)
%
Total accuracy
0.7033 (3)
0.7333 (1)
0.6667 (6)
0.6767 (5)
0.7000 (4)
0.7067 (2)
AUC
0.8490 (6)
0.8865 (1)
0.8645 (3)
0.8628 (4)
0.8710 (2)
0.8589 (5)
Type I accuracy
0.6052 (6)
0.6979 (1)
0.6127 (5)
0.6529 (2)
0.6481 (3)
0.6272 (4)
Type II accuracy
0.7454 (2)
0.7485 (1)
0.6898 (6)
0.7130 (5)
0.7222 (4)
0.7407 (3)
θ = 20%
Total accuracy
0.7533 (3.5)
0.7867 (1)
0.7633 (2)
0.7533 (3.5)
0.7500 (5)
0.7433 (6)
AUC
0.8764 (2.5)
0.8902 (1)
0.8493 (6)
0.8764 (2.5)
0.8643 (5)
0.8692 (4)
Type I accuracy
0.7138 (2.5)
0.7598 (1)
0.6526 (5)
0.7138 (2.5)
0.6712 (4)
0.6384 (6)
Type II accuracy
0.7703 (5.5)
0.7982 (2)
0.8108 (1)
0.7703 (5.5)
0.7838 (4)
0.7883 (3)
%
Total accuracy
0.6867 (3.5)
0.7097 (1)
0.6867 (3.5)
0.6667 (6)
0.6700 (5)
0.6933 (2)
AUC
0.8472 (4)
0.8625 (1)
0.8286 (5)
0.8491 (3)
0.8180 (6)
0.8524 (2)
Type I accuracy
0.5946 (5)
0.6375 (1)
0.5941 (6)
0.6016 (3)
0.5956 (4)
0.6054 (2)
Type II accuracy
0.7396 (2.5)
0.7406 (1)
0.7349 (4)
0.7031 (6)
0.7188 (5)
0.7396 (2.5)
%
Total accuracy
0.6733 (3)
0.6867 (1)
0.6740 (2)
0.6533 (6)
0.6600 (5)
0.6633 (4)
AUC
0.8172 (3)
0.8596 (1)
0.8484 (2)
0.8083 (5)
0.7908 (6)
0.8125 (4)
Type I accuracy
0.5832 (3)
0.6101 (1)
0.5639 (5)
0.5997 (2)
0.5736 (4)
0.5498 (6)
Type II accuracy
0.7071 (3)
0.7240 (1.5)
0.7240 (1.5)
0.6719 (6)
0.6919 (5)
0.6971 (4)
Average rank
3.63
1.13
4.00
3.83
4.40
4.03
Note: the bold-face in Table 3 shows the maximum of each row. The numbers in parentheses are the ranks of the six models with the corresponding evaluation criterion in each row.
