Use of machine learning techniques for identifying ischemic stroke instead of the rule-based methods: a nationwide population-based study

Table 2 AUROC, AUPRC, F₁ score, precision, recall, accuracy, and specificity of each model

Operational definitions	AUROC	AUPRC	F₁ score	Precision	Recall	Accuracy	Specificity
Rule-based method	91.47	73.39	68.68	54.80	91.98	91.47	91.41
Statistical models or tree-based machine learning techniques
Logistic	90.76	88.90	86.39	84.45	88.43	86.06	83.68
Random Forest	92.70	91.70	79.03	92.55	68.96	81.68	94.43
XGBoost	94.46	92.80	91.50	91.16	91.83	91.45	91.08
Neural network-based deep learning techniques: with embedding
MLP	98.00	98.00	94.82	95.41	94.24	94.85	95.45
LSTM	100.00	100.00	99.04	99.21	98.78	99.04	99.21
GRU	100.00	100.00	99.81	99.92	99.69	99.81	99.93
CNN	100.00	100.00	99.55	99.46	99.64	99.55	99.46

AUROC: the area under the receiver operating characteristic curve, AUPRC: the area under precision–recall curve, Rule-based method: individuals who are hospitalized with a diagnosis of I63 and have received anti-platelet therapy and anti-coagulant therapy within 30 days of diagnosis, LSTM long–short-term memory, CNN Convolutional Neural Networks, MLP Multi-Layer Perceptron, GRU Gated Recurrent Unit

ISSN: 2047-783X