Table 3.
Performance comparison among AdaBoost, Cost-sensitive decision tree, AdaCost, and our method RankCost with respect to F-measure (F), TPR (R), and PPV (P) values and and their 95 % confidence intervals
| AdaBoost | Cost-sensitive | AdaCost | RankCost | ||
|---|---|---|---|---|---|
| Decision tree | |||||
| Breast Cancer | Cost | 1:0.2 | 1:0.1 | ||
| F | 0.465 [0.217,0.713] | 0.468 [0.366,0.570] | 0.502 [0.358,0.646] | 0.494 [0.376, 0.612] | |
| R | 0.388 [0.104, 0.672] | 0.765 [0.423,1.107] | 0.906 [0.603, 1.209] | 0.494 [0.270, 0.718] | |
| P | 0.579 [0.152, 1.000] | 0.337 [0.263, 0.411] | 0.347 [0.211, 0.483] | 0.494 [0.392, 0.596] | |
| Hepatitis | Cost | 1:0.4 | 1:0.1 | ||
| F | 0.5 [0.002,0.998] | 0.603 [0.261, 0.945] | 0.628 [0.200,1.056] | 0.628 [0.280,0.976] | |
| R | 0.469 [-.215, 1.153] | 0.688 [0.212, 1.164] | 0.719 [0.165,1.273] | 0.843 [0.465,1.221] | |
| P | 0.536 [-.122,1.194] | 0.537 [0.047,1.022] | 0.561 [0.159, 0.963] | 0.500 [0.158,0.842] | |
| Diabetes | Cost | 1:0.4 | 1:0.5 | ||
| F | 0.595 [0.447,0.743] | 0.658 [0.472,0.844] | 0.661 [0.493,0.829] | 0.692 [0.532,0.852] | |
| R | 0.526 [0.308,0.744] | 0.757 [0.589,0.825] | 0.731 [0.527,0.935] | 0.802 [0.638,0.966] | |
| P | 0.684 [0.508,0.860] | 0.582 [0.400,0.764] | 0.603 [0.403,0,803] | 0.609 [0.431,0.787] | |
| Sick Euthyroid | Cost | 1:0.2 | 1:0.2 | ||
| F | 0.007 [-.033,0.047] | 0.645 [0.549,0.741] | 0.616 [0.538,0.694] | 0.612 [0.538,0.686] | |
| R | 0.003 [-.017,0.023] | 0.826 [0.714,0.938] | 0.785 [0.659,0.911] | 0.942 [0.814,1.070] | |
| P | 0.1 [-.500,0.700] | 0.53 [0.416,0.644] | 0.507 [0.429,0.585] | 0.453 [0.389,0.517] |
The cost settings are those with which AdaBoost and Cost-sensitive decision tree can achieve the best F-measure values.