Skip to main content
. 2022 Sep 21;5(3):e33460. doi: 10.2196/33460

Table 2.

Machine learning classification results of models trained on automatic transcripts compared to results of models trained on manually corrected transcripts.

Task and model type Automatic transcripts AUROCa Manually corrected transcripts AUROC Change in AUROCb
Picture description

RFc 0.617 0.687 0.070d

GNBe 0.662 0.725 0.063d

LRf 0.671 0.743 0.072d

BERTg 0.618 0.686 0.068d
Experience description

RF 0.503 0.636 0.133d

GNB 0.549 0.677 0.128d

LR 0.543 0.674 0.131d

BERT 0.630 0.650 0.020d

aAUROC: area under the receiver operating characteristic curve.

bPositive change in AUROC indicates that the manually corrected transcript model outperformed the automatic transcript model.

cRF: random forest.

dIndicates P<.001.

eGNB: Gaussian naive Bayes.

fLR: logistic regression.

gBERT: Bidirectional Encoder Representations from Transformers.