Predictive modelling for classification

I want to turn a large, well-organised spreadsheet of historical records into a reliable classifier. The goal is straightforward: feed in new rows of structured data and instantly receive a category prediction with clear confidence scores.

You will start with a clean CSV export that already includes labelled outcomes. Feel free to work in Python with scikit-learn, XGBoost, LightGBM, or a comparable library—whatever gets the best accuracy while keeping inference times low. I’m open to simple baseline models first, followed by feature engineering and hyper-parameter tuning to squeeze out extra performance.

Because this is strictly a classification task, success is measured by precision, recall, F1 and a well-calibrated ROC-AUC on a hold-out test set. I’d also like a brief explanation notebook so non-technical stakeholders can understand how key features influence the prediction.

Deliverables
• Clean, commented source code and environment file
• Trained model artefact (pickled or equivalent)
• Evaluation report with the metrics above and confusion matrix visuals
• Short Markdown or Jupyter notebook highlighting feature importance and usage instructions

If you can optionally add a lightweight REST endpoint (FastAPI or Flask) for real-time predictions, let me know—the extra polish would be appreciated.