🧠 Supervised, Semi-Supervised & Active Learning (DSCI 552 - HW8)

Perfect-score (100/100) project for Machine Learning (DSCI 552) @ USC.
This assignment compared supervised, semi-supervised, unsupervised, and active learning methods on real-world datasets.

---

🚀 Highlights

• Supervised: L1-penalized SVMs with normalized features + 5-fold CV.
• Semi-Supervised: Self-training with margin-based unlabeled selection.
• Unsupervised: K-Means & Spectral Clustering with majority voting.
• Active Learning: SVM uncertainty sampling vs Passive random sampling.
• Monte Carlo Simulation: 30–50 runs for stable error estimates.

---

📊 Results

• Supervised SVMs gave strongest baselines.
• Semi-Supervised improved when labeled data was scarce.
• Active Learning outperformed Passive — reaching low error with fewer labels.
• Unsupervised underperformed but revealed meaningful data structures.

---

📂 Files

• Chenyi_Weng_HW8.ipynb → Full code + analysis
• Homework8.pdf → Assignment description

---

👩‍💻 Author

Chenyi Weng | M.S. Spatial Data Science @ USC (Class of 2025)
🌐 Portfolio • 💼 LinkedIn • 📧 wengchen@usc.edu