NataVisionaries

Overview

This project builds a machine learning pipeline to predict the quality (OK / KO) of Pastel de Nata products for the Nata Visionaries brotherhood. The model will classify each production based on recipe and process data, saving pastries from being destroyed.

Project Structure / Notebooks

The project will consist of 5 Jupyter notebooks as required:

1. MLXX NB1 DATAEXPLORATION.ipynb

• Load learn.csv
• Understand data structure & feature types
• Summary statistics, distributions, correlations
• Initial observations & potential data issues

2. MLXX NB2 PREPROCESSING.ipynb

• Handle missing values / outliers
• Encode categorical features (origin)
• Scale / normalize numerical values if needed
• Save processed dataset (if needed for later notebooks)

3. MLXX NB3 FEATUREMGMT.ipynb

• Engineer new relevant features
• Drop irrelevant / redundant variables
• Justify transformations
• Export cleaned + engineered dataset

4. MLXX NB4 MODELLING.ipynb

• Train/test split
• Experiment with ML classifiers (e.g. RF, XGBoost, etc.)
• Compare accuracy
• Optional hyperparameter search

9. MLXX NB9 FINAL.ipynb

• Load raw data directly from files (learn.csv, predict.csv)
• Rebuild best data prep + feature steps
• Load best model configuration
• Train and export Kaggle sampred.csv-style predictions

Data Files

• learn.csv → full training dataset including target
• predict.csv → same features, no target, must predict
• sampred.csv → example submission format

Kaggle Submission

• Output must match sampred.csv format
• Submit early and often -> leaderboard feedback required

Deadlines

• Kaggle closes: 17 Dec
• Final notebook submission: 20 Dec (follow naming rules!)

Notes

• Markdown explanations in every notebook
• No data leakage
• NB1 and NB9 must run standalone from raw files
• Use accuracy as main metric

---

This README is a quick roadmap to keep the project organized and correctly aligned with grading rules.