Kaggle Playground Series S6E2 — Predicting Heart Disease

This is my first time using the XGBoost + CatBoost blend approach, but I wanted to give it a go and see if we can score higher on this exercise to find what features work best. It was difficult but fun!

Competition

• Task: Binary classification — predict presence/absence of heart disease
• Metric: AUC (Area Under ROC Curve)
• Data: 630K train rows, 270K test rows, 13 numeric features, no nulls
• Link: https://www.kaggle.com/competitions/playground-series-s6e2

Results

Leaderboard: #817 — Public LB Score: 0.95355

Model	OOF AUC
XGBoost (7-seed avg)	0.95542
CatBoost (7-seed avg)	0.95550
50/50 Blend	0.95550

Tuning Journey

Run	Learning Rate	Seeds	OOF AUC	Public LB	Rank
v1	0.05	3	0.95547	0.95353	#850
v2	0.02	7	0.95550	0.95355	#817

Project Structure

├── data.py          # Data loading and target encoding
├── features.py      # Feature column definitions and transformations
├── model.py         # Stratified K-fold CV with multi-seed averaging
├── submission.py    # Blend predictions and generate submission CSV
├── requirements.txt
└── data/
    ├── train.csv
    ├── test.csv
    └── sample_submission.csv

Usage

pip install -r requirements.txt
python submission.py