K-Means

K-Means is an algorithm cluster given sample data into k groups, and all sample will end up in some cluster. The algorithm initialize k center node randomly. For each iteration:

• For each sample, we find the nearest center node.
• For each group, we calculate the center node again using all samples.
• Repeat until group member converge.
• If a group loss all member, we drop it, may or may not re-initialize it randomly.

K-Means performs poorly when feature dimension goes high, or k goes high. And K-Means also makes the assumption that the cluster is round shaped, which leads to problem in many case.

DBSCAN

DBSCAN stands for “Density-Based Spatial Clustering of Applications with Noise”. As its name, it groups samples based on the density respecting parameter eps. And any sample not able to join a cluster with more than minPts sample, will be classify as noise. With DBSCAN, we do not have direct control over cluster number, and will have sample belongs to no cluster.

Agglomerative

Agglomerative Hierarchical Clustering starts with putting each sample into its own cluster. And repeatedly merge closest cluster until reaching desired number of cluster. The merging history shows the hierarchy (called a dendrogram).

Measure: Silhouette Score

The silhouette score is a metric used to evaluate the quality of a clustering result, regardless of clustering method.

• : average distance to all other points in the same cluster.
• : lowest average distance to points in any other cluster.
• silhouette score: