Table 3

Performance metrics of the machine learning models based on the PH data set

Machine learning algorithmBiochemical hypoglycemia (BG <4 mmol/L)Clinically significant hypoglycemia (BG <3 mmol/L)
AUROCPrecisionRecallAUROCPrecisionRecall
Logistic regression0.730.480.100.750.390.10
SGD0.740.120.100.770.100.10
kNN0.620.400.180.620.300.15
Decision tree0.810.700.710.840.680.73
Gaussian-naive Bayes0.810.470.680.860.330.81
Bernoulli-naive Bayes0.820.600.600.860.470.67
Multinomial-naive Bayes0.750.100.100.790.100.10
SVM0.790.730.100.830.410.10
QDA0.770.230.960.890.150.97
Random forest0.940.860.670.930.960.66
Extra trees0.930.850.680.930.940.66
LDA0.880.690.750.900.720.72
Passive aggressive0.760.460.250.770.330.10
AdaBoost0.890.680.600.930.630.46
Bagging0.930.840.700.920.930.67
Gradient boosting0.960.870.700.960.960.67
XGBoost0.960.880.700.960.970.67
MLP0.740.570.170.780.470.14
Mean (SD)0.82 (0.10)0.59 (0.25)0.49 (0.29)0.85 (0.10)0.55 (0.31)0.48 (0.31)
  • kNN, k-nearest neighbor; LDA, linear discriminant analysis; MLP, multilayer perceptron (artificial neural network); QDA, quadratic discriminant analysis; SGD, stochastic gradient descent; SVM, support vector machine.