Clustering Models

Clustering is an UnSupervised ML model that is used to group similar observations together into clusters based on some similarity metric like we use in KNN model.

It maintains the homogeneity inside the clusters and heterogeneity between the clusters.

• Major Algorithms for Clustering are:
  1. Partition Based Clustering
     • Hard Clustering (Non_Overlapping Clusters)
       • K-means (MiniBatch Variation is more popular)
         • Centroid based
         • NP-Hard problems
       • K-medoids
       • K-mode
       • Mean-Shift Clustering
     • Soft Clustering (Overlapping Clusters)
       • Fuzzy Clustering
         • C-means clustering
  2. Density Based Clustering
     • DBSCAN
     • OPTICS
  3. Hierarchal Clustering (Connectivity based clustering)
     • Divisive (Top-Down)
     • Agglomerative (Bottom-Up)
  4. Grid Based Clustering (Uses a multi-resolution grid data structure)
     • STING – explores statistical data stored in the grid cells
     • CLIQUE – defines a grid-and density-based approach for clustering in high-dimensional data space.
     • WaveCluster – clusters objects using a wavelet transform approach
  5. Distribution Based
     • The Expectation-Maximization (EM) algorithm 

K-Means Clustering is the most common and important clustering algorithm used in industry.

Here, 
K is the number of clusters, and 
Means is the Average of data points to form the specific cluster

MiniBatchKMeans is another variation of Standard K-Means, where we use only a Defined size of the Batch [i.e. Batch Size] for each iteration of K-Means clustering. This helps in faster convergence. 

This reduces the train/fit time but slightly increases the Error or WSS i.e. Inertia.