Embeddings Actuarial Loss Competition

Context

In the actuarial loss competition, we are provided with text data describing accident context and injury type.

Content

This is the first 20 components of a PCA performed on sentence embeddings of the claim descriptions. It is the concatenation of train and test data.

Acknowledgements

The embeddings were obtained with paraphrase distil roberta from sentence-transformers : https://github.com/UKPLab/sentence-transformers.

Inspiration

How can these embeddings explain the claim cost, and help predict it ?