Clustering

Algorithms

• K-Means
• DBSCAN
• To reduce dimensions
  • PCA
  • t-SNE

Evaluation

• High intra-cluster similarity, low inter-cluster similarity
• External
  • Precision, Recall
  • F1 score
• Internal Consistency
  • Sum of Square Error (SSE)
  • BetaCV, compactness and separability, average between intra-cluster and inter-cluster distance
• Silhouette Coefficient
  • Cohesion and separation
  • ranges from $-1$ to $1$
  • Calculate average coefficient of all points, $0.7$ is strong, $0.5$ is reasonable, $0.2$ is poor
  • $a(i)$ = mean distance with all points in the same cluster, $b(i)$ = mean distance with all points in the nearest other cluster

$$s(i)=\frac{b(i)-a(i)}{\max (a,b)}$$