Table 1 (abstract P153). Experimental results (superslow bleeding subjects)
From: 39th International Symposium on Intensive Care and Emergency Medicine
Model | AUC | AUC@FPR<1% | TPR@FPR=0.1% | TNR@FNR=1% |
|---|---|---|---|---|
Support Vector Machine | 0.8936 | 0.5306 | 0.0132 ± 0.0012 | 0.0631 ± 0.0038 |
Logistic Regression | 0.8445 | 0.5132 | 0.0062 ± 0.0014 | 0.0484 ± 0.0083 |
Naive Recurrent Neural Network (nRNN) | 0.9015 | 0.6077 | 0.0439 ± 0.2558 | 0.0583 ± 0.2875 |
RF on statistical features (baseline) | 0.9705 | 0.6386 | 0.1456 ± 0.3242 | 0.6386 ± 0.1972 |
Long Short-Term Memory (LSTM) | 0.9263 | 0.7010 | 0.3289 ± 0.1357 | 0.0981 ± 0.3140 |
Gated Recurrent Unit (GRU) | 0.9449 | 0.7469 | 0.3832 ± 0.2267 | 0.2227 ± 0.2881 |
Dilated, causal convolution | 0.9360 | 0.5390 | 0.0163 ± 0.3763 | 0.1564 ± 0.1991 |