✈️ Weather & Flight Performance Prediction

DATASCI 261 – Machine Learning at Scale | Team 1_3
Ahsin Saleem · Anshul Zutshi · Cat Weiss · Jasmol Dhesi · Umesh Kant

Predict whether a flight will be delayed by 15 minutes or more using historical flight, weather, and airport data (2015–2019).
The project maximizes recall to avoid missing potential delays and empowers travelers, airlines, and airports with actionable insights.

---

📊 Key Results

Metric	Single‑Layer MLP
Precision	0.8232
Recall	0.6961
AUC	0.8992
F1	0.7544

Data leakage issues were investigated and resolved, ensuring reliable evaluation.

---

🔍 Highlights

• Implemented Logistic Regression, Random Forest, and Multilayer Perceptron
• Extensive hyperparameter tuning + Bayes‑rule calibration for down‑sampling bias
• Feature engineering: rolling three‑month delay counts, holiday flags, weather condition indicators
• Handled ~31 M flight rows & ~630 M weather rows with PySpark on Databricks

---

📁 Repository Structure

📦 project-root/
├── Phase 3 – Final Report‑Team‑1‑3.ipynb # EDA + early modeling
├── Phase 3 – Final ‑ Team‑1‑3.ipynb # Final modeling notebook
├── 261 Team 1_3 Project Report Final.pptx # Slide deck
└── README.md

---

🛠 Tech Stack

Python · PySpark · Spark MLlib · Databricks · AWS
NumPy · Pandas · Matplotlib · Seaborn

---

🚀 Next Steps

1. Add Decision Trees for interpretability
2. Explore ensemble methods (boosting, stacking)
3. Further hyperparameter sweeps for marginal gains

---

🤝 Contributing

Pull requests are welcome. Please open an issue to discuss proposed changes before submitting.

---

📜 License

Distributed under the MIT License. See LICENSE for details.