Contributions are welcome, create an issue or pull request to improve the project.
The current setup constrains the problem to variants with multiple submissions. It might be interesting to predict which variants with single submissions will have conflicting classifications for future releases of ClinVar. If there are other ways to apply machine learning to predict relevant classifications from ClinVar please suggest the idea!