. 2021 Aug 17;11:16682. doi: 10.1038/s41598-021-95957-w

Table 2.

The comparison results between DF-COVID-19 and other deep learning models.

Model	Accuracy	AUC	Precision	Sensitivity	F1-score
ANN in¹³ train-test split	86.90%	85.0%	87.13%	87.13%	87.13%
10 fold cross-validation	86.0%	56.15%	88.55%	95.78%	91.34%
CNN in¹³ train-test split	87.35%	80.0%	88.47%	88.67%	88.56%
10 fold cross-validation	88.0%	61.49%	89.48%	92.48%	90.38%
CNNLSTM in¹³ train-test split	92.30%	90.0%	92.35%	93.68%	93.0%
10 fold cross-validation	84.16%	58.89%	89.26%	92.14%	90.01%
CNNRNN in¹³ train-test split	86.24%	69.0%	87.55%	87.55%	87.55%
10 fold cross-validation	85.66%	64.08%	89.77%	94.23%	91.20%
LSTM in¹³ train-test split	90.34%	83.0%	89.97%	89.98%	89.97%
10 fold cross-validation	86.66%	62.50%	86.75%	99.42%	91.89%
RNN in¹³ train-test split	84.0%	83.0%	84.28%	84.27%	84.27%
10 fold cross-validation	84.16%	52.45%	87.83%	96.04%	90.61%
DF-COVID-19 with¹³ features	93.98% [95%CI 88.29–100]	94.91% [95%CI 84.1–99.8]	85.66% [95%CI 56.3–100]	66.3% [95%CI 35.4–92.3]	73.33% [95%CI 48.5–92.0]