Skip to main content

Table 4 Accuracy and area under the ROC curve (AUC) results for bag of words (BOW)-based models and the word embedding-based deep learning models along with 95% confidence intervals (CI)

From: Automated detection of altered mental status in emergency department clinical notes: a deep learning approach

Category Modela AUC (95% CI) Accuracy Epochs
BOW models RF 0.975 (0.967–0.983) 0.921 N/A
LASS 0.973 (0.964–0.982) 0.912 N/A
SVM 0.967 (0.957–0.976) 0.912 N/A
MLP 0.947 (0.934–0.960) 0.883 N/A
SDT 0.934 (0.918–0.950) 0.911 N/A
NBC 0.924 (0.908–0.940) 0.838 N/A
Deep learning models CNN_D200 0.985 (0.979–0.992) 0.945 30.8
CNN_W2V 0.985 (0.979–0.991) 0.942 25.0
CNN_D50 0.984 (0.978–0.991) 0.944 36.6
  1. aModel abbreviations are described in the text
  2. The number of epochs for training the deep learning is based on the early stopping condition as described in the methods. The entries are sorted in descending order of AUC within each category. Bolding indicates results for the best performing models
Back to article page