Clustering Techniques Overview

Clustering Techniques

• Clustering groups objects based on their similarities.
• Clustering algorithms are categorized into:
  • Partitioning methods
  • Hierarchical methods
  • Density-based methods
  • Grid-based methods
  • Model-based methods
• K-means and K-medoids are popular partitioning methods.
• AGNES, DIANA, Birch, and Chameleon are common hierarchical methods.
• DBSCAN, OPTICS, and DENCLUE are density-based techniques.
• STING and CLIQUE are grid-based methods, with CLIQUE specifically for subspace clustering.

Distance Measures

• Single linkage: The smallest distance between elements in different clusters.
• Complete linkage: The largest distance between elements in different clusters.
• Average linkage: The average distance between all pairs of elements in different clusters.
• Centroid: The distance between the centroids (means) of two clusters.
• Medoid: The distance between the medoids of two clusters.

Cluster Measures

• Centroid: The center of a cluster.
• Radius: The square root of the average distance from any data point to the centroid.
• Diameter: The square root of the average mean squared distance between all pairs of points.

Clustering Validation

• The Hopkins statistic measures cluster tendency; values of >0.5 suggest clustering potential.
• The Davies-Bouldin Index assesses the quality of clustering, with lower values indicating better clustering. The DB value is based on the ratio of intra-cluster variance to the inter-cluster distance.
• The Dunn Index also measures clustering quality. This is calculated by taking the minimum inter-cluster distance divided by the maximum intra-cluster distance. A higher Dunn Index indicates better-quality clustering.
• Silhouette coefficient: Provides a measure of how similar a data point is to its own cluster compared to other clusters.

Hierarchical Clustering (AGNES and DIANA)

• AGNES (Agglomerative Nesting): A bottom-up hierarchical clustering method where clusters begin as individual data points and iteratively merge the most similar clusters.
• DIANA (Divisive Analysis): A top-down hierarchical clustering method where a single cluster is initially formed, and then it repeatedly splits the cluster that has the largest average intra-cluster distance.

Density-Based Clustering (DBSCAN and OPTICS)

• DBSCAN (Density-Based Spatial Clustering of Applications with Noise): Finds dense clusters of arbitrary shapes by considering local density.
• OPTICS (Ordering Points To Identify the Clustering Structure): Also a density-based method similar to DBSCAN, ordering the data points to identify cluster structure by core-distance and reachability-distance.

K-Means Variations

• PAM (Partitioning Around Medoids): Robust to outliers compared to K-means, since it uses medoids instead of means.

• CLARA (Clustering LARge Applications): An improved algorithm of PAM, used for large dataset, samples data to perform clustering.

• CLARANS (Clustering Large Applications based on Randomized Search): Improves on earlier sampling-based methods, more sophisticated in handling outliers and clusters of various sizes.

This quiz covers various clustering techniques used in data analysis. It explores partitioning methods like K-means, hierarchical methods such as AGNES, and density-based techniques including DBSCAN. Additionally, the quiz discusses distance measures important for cluster formation and evaluation.