Table 2 Generalization performance of classifiers with optimized parameters, presented as confusion matrices and balanced accuracy ± standard deviation across five-fold cross-validation

Linear discriminant analysis

N/A

Actual: No Unplanned Visits

50.8% ± 1.4%

49.2% ± 1.4%

Actual: ≥ 1 Unplanned Visit

24.8% ± 1.0%

75.2% ± 1.0%

Average

63.0% ± 0.7%

Quadratic discriminant analysis

N/A

Actual: No Unplanned Visits

82.5% ± 0.6%

17.5% ± 0.6%

Actual: ≥ 1 Unplanned Visit

56.3% ± 0.8%

43.7% ± 0.8%

Average

63.3% ± 0.4%

Linear SVM

Cost = 25

Actual: No Unplanned Visits

71.1% ± 0.8%

28.9% ± 0.8%

Actual: ≥ 1 Unplanned Visit

39.8% ± 1.0%

60.2% ± 1.0%

Average

65.7% ± 0.3%

Radial SVM

Cost = 50;

Gamma = 0.1

Actual: No Unplanned Visits

57.6% ± 1.4%

42.5% ± 1.4%

Actual: ≥ 1 Unplanned Visit

28.4% ± 0.9%

71.6% ± 0.9%

Average

64.6% ± 0.8%

Single hidden layer NN

Hidden layer = 20 nodes;

Iterations = 200;

Decay = 0.0

Actual: No Unplanned Visits

50.7% ± 28.7%

49.3% ± 28.7%

Actual: ≥ 1 Unplanned Visit

31.6% ± 20.4%

68.4% ± 20.4%

Average

59.5% ± 7.7%

Triple hidden layer DNN

Hidden layers = 20 nodes;

Learning = 1.0;

Momentum = 0.5;

Iterations = 20

Actual: No Unplanned Visits

65.7% ± 14.5%

34.4% ± 14.5%

Actual: ≥ 1 Unplanned Visit

36.7% ± 14.6%

63.3% ± 14.6%

Average

64.5% ± 0.8%

XG boost

Max depth = 20; Eta = 0.90; # rounds = 200; Gamma = 10; Min. child weight = 10; Ratio of column per tree = 1.0

Actual: No Unplanned Visits

33.9% ± 30.8%

66.1% ± 30.8%

Actual: ≥ 1 Unplanned Visit

16.7% ± 15.3%

83.3% ± 15.3%

Average

58.6% ± 7.8%

Logistic Regression

N/A

Actual: No Unplanned Visits

60.4% ± 0.8%

39.6% ± 0.8%

Actual: ≥ 1 Unplanned Visit

29.8% ± 0.8%

70.2% ± 0.8%

Average

65.3% ± 0.7%