Data analytics and clinical feature ranking of medical records of patients with sepsis

Table 3 Septic shock (vasopressors) prediction

method	MCC	F₁ score	accuracy	TP rate	TN rate
RF	0.32±0.14	0.88±0.03	0.80±0.04	0.88±0.04	0.43±0.15
MLP	0.31±0.13	0.87±0.03	0.79±0.04	0.87±0.04	0.47±0.15
LR	0.31±0.13	0.84±0.04	0.76±0.05	0.79±0.06	0.62±0.15
DL	0.30±0.11	0.83±0.05	0.73±0.04	0.78±0.05	0.61±0.16
NB	0.27±0.08	0.79±0.09	0.70±0.09	0.72±0.15	0.59±0.18
SVM (linear)	0.26±0.13	0.82±0.06	0.75±0.06	0.82±0.09	0.49±0.18
k-NN	0.23±0.14	0.81±0.06	0.71±0.07	0.76±0.10	0.50±0.20
SVM (kernel)	0.22±0.13	0.79±0.06	0.70±0.06	0.75±0.09	0.50±0.18
DT	0.18±0.13	0.78±0.06	0.67±0.07	0.72±0.10	0.50±0.19
method	PR AUC	ROC AUC	PPV	NPV
RF	0.18±0.07	0.28±0.29	0.47±0.16	0.87±0.05
MLP	0.16±0.05	0.28±0.28	0.46±0.16	0.87±0.05
LR	0.11±0.04	0.26±0.31	0.41±0.12	0.90±0.04
DL	0.11±0.04	0.26±0.31	0.39±0.12	0.88±0.04
NB	0.11±0.06	0.26±0.29	0.34±0.09	0.89±0.04
SVM (linear)	0.15±0.05	0.26±0.24	0.37±0.13	0.86±0.06
k-NN	0.13±0.06	0.25±0.30	0.33±0.12	0.87±0.06
SVM (kernel)	0.13±0.05	0.24±0.20	0.32±0.11	0.86±0.06
DT	0.12±0.05	0.23±0.29	0.29±0.10	0.86±0.06

Performance of the learned models with the different methods evaluated with the different metrics, expressed in the format “average value ± standard deviation”, obtained on 100 executions. DT: decision tree. MLP: multi-layer perceptron neural network. RF: random forest. k-NN: k-nearest neighbors. DL: deep neural network with 3 hidden layers and weight decay. LR: logistic regression. NB: Naïve Bayes. SVM (kernel): support vector machine with kernel. SVM (linear): linear support vector machine. MCC: Matthews correlation coefficient. TP rate: true positive rate (sensitivity, recall). TN rate: true negative rate (specificity). PR: precision-recall curve. ROC: receiver operating characteristic. AUC: area under the curve. MCC: worst value –1.00 and best value +1.00. PPV: positive predictive value (precision). NPV: negative predictive value. F₁ score, accuracy, TP rate, TN rate, PR AUC, ROC AUC, PPV, NPV: worst value 0.00 and best value 1.00. Imbalance of this dataset: yes septic shock class: 1’s positives, #elements = 67 (18.41%), and no septic shock class: 0’s negatives, #elements = 297 (81.59%)

ISSN: 1756-0381