Automatic Prediction of Recurrence of Major Cardiovascular Events: A Text Mining Study Using Chest X-Ray Reports

Table 5

Performance comparison of different experimental scenarios using AUC and misclassification rate when clinical predictors are not available.

Classifier	AUC	Misclassification rate
D-	0.685	0.242
D-SVM	0.572	0.246
D-NN	0.567	0.214
DB-	0.703	0.247
DB-SVM	0.674	0.163
DC-	0.705	0.239
DC-SVM	0.534	0.247
D-MI-	0.708	0.235
D-MI-SVM	0.568	0.247
D-MI-CNN	0.667	0.228
D-MI-LSTM	0.708	0.209
D-MI-BiLSTM	0.745	0.204

^aLR trained on demographic variables. ^bLR trained on demographic variables and BOW representation. ^cLR trained on demographic variables and clustering-based representation. ^dMultimodal learning LR trained on demographic variables and word embeddings.