Skip to main content
. 2024 Sep 14;7(3):ooae084. doi: 10.1093/jamiaopen/ooae084

Figure 4.

Graphs summarizing the prediction results of different methods on 942 phenotypes in the MIMIC-III database, showing the advantage of PheW2P2V, which has bigger improvements over baseline methods when phenotypes are rare.

Medians, 25th and 75th percentiles of AUC-ROC, max F1-score, and AUC-PR across binned 300 phenotypes with descending prevalence in the MIMIC-III databases for the proposed PheW2P2V, the LASSO regression, the random forest classifier, the gradient boosted tree classifier, and the unweighted P2V.