From: Improving rare disease classification using imperfect knowledge graph
Percentage Bins | (0, 0.02%] | (0.02%, 0.05%] | (0.05%, 0.1%] | (0.1%, 0.5%] | (0.5%, 1%] | |||||
---|---|---|---|---|---|---|---|---|---|---|
5 diseases | 3 diseases | 2 diseases | 7 diseases | 9 diseases | ||||||
F1 | MRR | F1 | MRR | F1 | MRR | F1 | MRR | F1 | MRR | |
BOW | 91.58 | 93.36 | 29.76 | 53.97 | 90.49 | 93.49 | 88.69 | 92.64 | 92.6 | 95.09 |
LSTM | 0.00 | 4.03 | 0.00 | 4.75 | 0.00 | 9.64 | 22.38 | 44.68 | 85.86 | 93.55 |
UpSample | 88.36 | 94.81 | 52.22 | 66.54 | 90.11 | 93.06 | 89.36 | 94.27 | 92.62 | 95.76 |
χ2 | 91.38 | 95.83 ∗ | 47.97 | 65.12 | 90.40 | 93.68 | 91.92 | 95.41 | 93.84 | 96.45 |
BOW+ χ2 | 93.37 ∗ | 97.55 ∗ | 42.14 ∗ | 62.80 ∗ | 90.73 | 93.95 | 92.01 | 95.55 | 94.05 | 96.43 |
KG1 | 91.06 | 97.47 ∗ | 22.63 | 43.64 | 48.52 | 48.11 | 80.54 | 86.67 | 74.32 | 77.33 |
KG12 | 92.26 ∗ | 97.70 ∗ | 31.20 | 43.91 | 85.61 | 91.42 | 83.71 | 87.96 | 80.05 | 83.18 |
BOW+KG\(^{\text {pseudo-doc}}_{1}\) | 75.68 | 82.49 | 34.86 | 52.08 | 83.20 | 87.84 | 78.79 | 85.57 | 88.34 | 91.86 |
BOW+KG\(^{\text {pseudo-count}}_{1}\) | 88.14 | 91.02 | 30.04 ∗ | 52.62 | 89.02 | 93.61 | 85.54 | 88.64 | 90.8 | 93.34 |
BOW+KG\(^{\text {late-fusion}}_{1}\) | 89.01 | 95.41 ∗ | 29.76 | 48.8 | 68.63 | 70.80 | 86.18 | 89.65 | 86.89 | 86.21 |
BOW+KG\(^{\text {early-fusion}}_{1}\) | 92.30 ∗ | 97.66 ∗ | 54.73 ∗ | 69.88 | 90.27 | 92.54 | 91.00 | 95.05 | 93.59 | 95.92 |
BOW+KG\(^{\text {early-fusion}}_{12}\) | 93.43 ∗ | 97.13 | 47.78 | 62.04 | 91.68 | 95.41 | 90.70 | 94.49 | 93.46 | 95.70 |