Clustering is grouping collections of unlabeled data into a number of clusters based on similarity of data items.

Clustering Methods

• Centroid-based Clustering: Is a non-hierarchical clustering method where centroids for a specific number of clusters is defined and distance to it is used to group data items.
  • Algorithms
    • K-Means
    • Mini-Batch K-means
• Distribution-based Clustering: This method is used in data which is composed of distributions; where the distance from the distribution's center indicates the probability of item belonging to the distribution.
  • Algorithms
    • Gaussian mixture model (GMM)
• Density Method: It identifies and groups data points in areas of high concentrations together, assuming that they have more similarities and differences than points in a lower dense region.
  This method can take advantage of Kernel Density Estimation(KDE), also called Probability Density Function(PDF), to estimate the underlying distribution of data.
  • ✔️ this method has a good accuracy
  • ✔️ It has the ability to merge clusters
  • ✔️ Creates arbitrary-shaped distributions for dense areas
  • ✔️ It’s able to find outliers.
  • ❌ Is weak high dimensional data
  • Algorithms
    • DBSCAN
    • OPTICS
    • Mean-Shift
• Hierarchical Method: It forms first clusters in a tree-type structure, then creating new clusters from previously formed clusters.
  • Algorithms
    • BIRCH
    • CURE
    • Agglomerative Hierarchy clustering algorithm
• Partitioning Method: It partitions the objects into k clusters and each partition forms one cluster.
  • Algorithms
    • CLARANS
• Grid-based Method: formulates the data into a finite number of cells that form a grid-like structure.
  • Algorithms
    • CLIQUE
    • STING
• Graph-Based Methods: Utilizes graph Theory and treats data items as nodes and their connections in their edges as a measure of similarity.
  • Algorithms
    • Spectral Clustering

Resources:

• Google Machine Learning: Clustering