F1 Pit Stop Prediction | Kaggle Playground Series S6E5

Predicting whether an F1 driver will pit on the next lap using race strategy data.

---

🏁 Project Overview

• Competition: Kaggle Playground Series S6E5 (Predicting F1 Pit Stops)
• Task: Binary classification (pit or not on next lap)
• Metric: ROC AUC
• Public Score: 0.94541
• Rank: 2047/3023 (top 68%)

---

📊 Dataset

• 439,140 training samples, 188,165 test samples
• 15 features including:
  • Race context (lap, race number, position)
  • Tyre information (compound, age, wear)
  • Driver information
• Target imbalance: 80% no-pit, 20% pit

---

🔧 Approach

©️🖼️ Diagram Credits: https://gitdiagram.com/

Data Preprocessing

• Categorical encoding: Label encoding for Driver, Compound, Race
• Train/validation split: 80/20 stratified
• No missing values

Model

• Algorithm: LightGBM (gradient boosting with leaf-wise tree growth)
• Hyperparameters:
  • num_leaves: 31
  • learning_rate: 0.05
  • feature_fraction: 0.9
  • num_boost_round: 500
• Validation AUC: 0.9466

Results

• Public LB AUC: 0.94541
• Rank: 2047/3023 with single submission
• No feature engineering or hyperparameter tuning

---

📂 Files

• notebooks/baseline_lightgbm.ipynb — Main notebook with full pipeline
• submission.csv — Final predictions for test set
• README.md — This file

🛠 Setup

pip install pandas numpy scikit-learn lightgbm

🚀 Future Work

• Feature engineering (lap phase, stint features, position changes)
• Cross-validation (5-fold)
• Hyperparameter tuning
• Ensemble with XGBoost/CatBoost
• Threshold calibration

📚 References

• Kaggle Competition
• LightGBM Documentation

📬 Contact

Feel free to reach out for questions or feedback!