Table 11 Prediction model results for PE on UK test set (F1-w is the weighted F1 score).

From: At-admission prediction of mortality and pulmonary embolism in an international cohort of hospitalised patients with COVID-19 using statistical and machine learning methods

Models

Validation AUC

AUC

Accuracy

F1-w

Sensitivity

No undersampling

No threshold

Logistic Regression

72.5

69.4

64.9

77.3

65.5

LDA

72.2

69.2

98.4

97.6

0.0

Naive Bayes

70.4

67.2

98.3

97.6

0.5

Random forest

73.6

71.2

65.6

77.8

66.0

Stacking ensemble

63.0

65.7

66.1

78.2

65.2

Ensemble

73.0

70.3

64.7

77.1

67.0

Ensemble (XGBoost)

73.6

71.6

64.9

77.3

66.1

XGBoost

75.6

74.5

73.4

83.2

63.5

No undersampling

With threshold

Logistic regression

72.5

65.3

63.9

76.5

66.8

LDA

72.2

65.3

68.7

80.0

61.9

Naive Bayes

70.4

63.6

61.9

75.0

65.5

Random forest

73.6

64.5

71.9

82.2

56.9

XGBoost

73.8

65.5

68.5

79.9

62.5

With undersampling

No threshold

Logistic Regression

72.4

69.3

64.8

77.2

65.2

LDA

72.2

69.2

97.0

97.0

4.8

Naive Bayes

70.4

67.2

83.3

89.5

29.7

Random forest

74.3

71.7

68.8

80.1

62.4

Stacking Ensemble

64.5

65.9

67.5

79.2

64.3

XGBoost

73.8

71.7

66.8

78.6

65.1

With undersampling

With threshold

Logistic Regression

72.4

64.8

65.6

77.8

64.0

LDA

72.2

65.5

64.5

77.0

66.6

Naive Bayes

70.4

63.5

61.6

74.8

65.5

Random forest

74.3

65.5

69.3

80.5

61.6

XGBoost

73.8

65.9

62.8

75.7

69.1

  1. Best performing model performance values are in bold.