From: O2 supplementation disambiguation in clinical narratives to support retrospective COVID-19 studies
Classifier | Parameters - Values |
---|---|
SVM | vectorisor - TF-IDF vectorisor (in a range of 2450 to 2520 dimensions) |
kernel - rbf (Radial Basis Function) kernel | |
regularisation parameter - C value of 10 | |
cross-validation - 5 fold | |
RF | vectorisor - TF-IDF vectorisor (in a range of 2450 to 2520 dimensions) |
maximum features- square root of the total number of features | |
number of decision trees - 100 | |
cross-validation - 5 fold | |
LSTM & Bi-LSTM | Embedding layer - 300 dimensional |
LSTM / Bi-LSTM layer - 128 nodes | |
dropout and recurrent dropoutlayer -probability of 0.2 | |
dense output layer - 1 node | |
activation layer - sigmoid | |
cross-validation - 5 fold | |
loss - binary cross entropy | |
optimizer - adamax | |
CNN | Embedding layer - 300 dimensional |
1D convolutional layer with: | |
- filters - 256 | |
- window size - 5 | |
- activation layer - relu | |
dropout layer - probability of 0.5 | |
dense output layer - 1 node | |
activation layer - sigmoid | |
cross-validation - 5 fold | |
loss - binary cross entropy | |
optimizer - adamax |