Table 2 Performance comparison of different assessment models on the test set.

From: Automatic assessment of adverse drug reaction reports with interactive visual exploration

Measurements

ACA model

MLR

DT

Accuracy

\(85.99\%\)

\(80.82\%\)

\(85.39\%\)

\(F_1\)-2 (conditional/unclassified)

0.4

0

0

\(F_1\)-3 (possible)

0.8329

0.7716

0.8346

\(F_1\)-4 (probable/likely)

0.8775

0.8482

0.8692

\(F_1\)-5 (certain)

0.9083

0

0.8559

AUC-macro

0.9572

0.8603

0.9440

AUC-micro

0.9748

0.9572

0.9779

AUC-2 (conditional/unclassified)

0.9878

0.9580

0.9416

AUC-3 (possible)

0.9324

0.9162

0.9398

AUC-4 (probable/likely)

0.9291

0.9126

0.9176

AUC-5 (certain)

0.9796

0.6547

0.9768

  1. ACA model Automatic casualty assessment model, MLR Multinomial logistic regression, DT Decision tree, \(F_1\)-x: the \(F_1\) score of category x (category 2–5), AUC Area under curve, AUC-x the multiclass AUC of category x (category 2–5).
Back to article page