ffsshhttiikk What I do
- Discover hidden patterns in data
- Group similar observations without labels
- Reduce feature dimensions
- Find outliers and anomalies
- Learn latent representations
- Generate synthetic data
When to use me
Use me when:
- Labels are unavailable or expensive
- Exploring unknown data patterns
- Reducing data complexity
- Preprocessing for supervised learning
- Real-time clustering needed
Key Concepts
Clustering
- K-Means: Centroid-based, assumes spherical clusters
- Hierarchical: Dendrogram-based, any distance
- DBSCAN: Density-based, finds arbitrary shapes
- GMM: Probabilistic, soft clustering
Dimensionality Reduction
- PCA: Linear, variance-based
- t-SNE: Non-linear, visualization
- UMAP: Non-linear, structure preservation
- Autoencoders: Non-linear, learned
Evaluation
- Silhouette score
- Davies-Bouldin index
- Calinski-Harabasz index
- Reconstruction error (for autoencoders)