we made the classifier of the etiology for pleural effusion. Five kinds of the model were tested as a classifier:
- random-forest model
- gradient-boost model
- multinomial logistic regression model
- multi-layer perceptron using cross-entropy loss
- multi-layer perceptron using contrastive loss
the embedding space of the contrastive-loss model shows clustered data points by each category because the contrastive loss pulls the data of the same labels. With t-SNE and UMAP visualization, the embedding space is visualized.
You can see the embedding space and model metrics in our uploaded files.
Detailed methods and results will be published in the medical journal soon.