kishuvansh/MULE_Hunter_for_banks ? reverse-engineered prompt

Build me a complete machine learning project that detects money mule bank accounts from transaction data, based on the ideas in this repo.

I want it to take large banking transaction files, engineer the same kind of risk signals from the README, train a strong model, and then produce predictions for suspicious accounts along with the likely suspicious date window. Please keep the workflow practical, so I can run the main training notebook, see the feature generation, model training, evaluation, and a final submission file or output table. Also include the model explanation visuals like the SHAP charts if possible, and make the simple HTML page show the results in a clean way.

If you need to check current docs or best practices for the libraries involved, go ahead and look them up online. Make the finished project easy to understand and run from the notebooks in this repo, with clear outputs and sensible defaults.