Table 6 Discrimination of LightGBM models by input modality.
From: Multimodal machine learning for 5-year mortality prediction after percutaneous coronary intervention
Input modalities | AUC-ROC | PR-AUC | F1macro |
|---|---|---|---|
Tabular only | 0.789 (0.76–0.82) | 0.437 (0.39–0.48) | 0.657 (0.63–0.68) |
Visual only | 0.682 (0.64–0.72) | 0.297 (0.25–0.35) | 0.594 (0.55–0.64) |
Text only | 0.652 (0.60–0.70) | 0.246 (0.21–0.29) | 0.476 (0.43–0.52) |
Tabular + Visual | 0.810 (0.78–0.84) | 0.458 (0.41–0.51) | 0.662 (0.63–0.69) |
Tabular + Text | 0.802 (0.77–0.83) | 0.463 (0.41–0.52) | 0.674 (0.66–0.72) |
Visual + Text | 0.708 (0.66–0.75) | 0.294 (0.25–0.35) | 0.594 (0.55–0.64) |
All three | 0.814 (0.79–0.84) | 0.472 (0.42–0.52) | 0.682 (0.65–0.71) |
