Table 2. The factor importance for each sub-classifier. For each of them, 10 features which contribute most to the model are presented and evaluated. “tf” and “mf” stand for the TCM and MM feature respectively, and the numbers in brackets are standard deviations. 1) the value of variable “tongue color” from 1 to 6 represent pale, light red, red, crimson, light purple and dark purple respectively; 2) the value of “tongue fur color” from 1 to 6 represent white, yellowish, yellow, deep yellow, grey and dark. 3) For the binary attribute of TCM, if the value is 1, it means the patient has such symptom, and 0 otherwise. 4) For the categorical attributes of TCM, taking feeble pulse/replete pulse as an example, −1 for feeble pulse, 1 for replete pulse and 0 for normal control.
Classifier 1 (Healthy vs. CHB) | |||||
---|---|---|---|---|---|
Rank | Feature | Name | Healthy | CHB | P-value |
1 | tf_102 | Wiry pulse | 0.10 (0.31) | 0.66 (0.47) | 7.123e−11 |
2 | tf_44 | Lack of strength | 1.03 (0.19) | 1.49 (0.56) | 6.325e−16 |
3 | mf_5 | AST (IU/L) | 19.77 (5.06) | 43.48 (31.26) | <2.2e−16 |
4 | tf_100 | Feeble pulse/replete pulse | 0.17 (0.38) | 0.01 (0.13) | 0.031 |
5 | tf_80 | Tongue color | 2.31 (0.97) | 2.78 (1.17) | 0.03576 |
6 | tf_95 | Tongue coating color | 1.24 (0.51) | 1.88 (0.84) | 2.279e−07 |
7 | mf_4 | ALT (IU/L) | 19.8 (7.68) | 79.77 (89.4) | <2.2e−16 |
8 | mf_15 | HBsAg (IU/L) | 0.01 (<0.01) | 233.29 (55.8) | <2.2e−16 |
9 | tf_98 | Floating pulse/Deep pulse | 0.24 (0.44) | 0.06 (0.28) | 0.03535 |
10 | mf_86 | Mean platelet volume (fL) | 8.17 (0.96) | 10.43 (1.56) | 5.895e−14 |