juniorcl/transaction-fraud-detection ? reverse-engineered prompt

Build me a complete fraud detection project that predicts whether a transaction is fraud or not, using the transaction data in the repo. I want it to feel like a real data science portfolio project, with a clear notebook that explores the data, cleans it, creates a few useful features, tests some simple business ideas, trains a few machine learning models, and compares them using sensible fraud metrics like balanced accuracy, precision, recall, and F1.

Please include a final model that does the best job on fraud detection, save the trained model and any helper functions, and make it possible to use the model from a small Flask API. The app should accept transaction details and return a clear fraud or not fraud prediction. Keep the code organized, easy to follow, and make the notebook readable with plots and short explanations. If you need to look up current best practices or docs, go ahead.