CelebA Bald vs Non-Bald Anomaly Detection

This project explores anomaly detection techniques on a subset of the CelebA dataset, where the target anomaly is the presence of bald individuals.

We treat "bald" as the rare event and apply unsupervised models such as Isolation Forest and One-Class SVM to detect it.

---

Dataset

• Source: ADRepository - celeba_baldvsnonbald.arff

---

Goals

• Preprocess categorical data
• Apply anomaly detection models
• Evaluate model performance on detecting the bald class