Data Clustering Assignment Help

COMPONENTS OF CLUSTERING

Pattern representation – It refers to the number of classes, number of available patterns, the number, type and scale of the features available to the clustering algorithm.

• Pattern proximity measure to the data domain – Pattern proximity is usually measured by the distance function presented in pairs or patterns.
• Clustering or grouping – The grouping system can be performed in many ways –
• Hierarchical clustering – It acts as an algorithm in which the similar clusters are split or merged between the series of partitions.
• Partition clustering – It is an algorithm that optimizes a clustering criterion.
• Other methods are probabilistic and graph – theoretic clustering methods.
• Data abstraction – It includes the extraction of simple and compact representation of a data set. The typical data abstraction includes compact description of each cluster in terms of cluster prototypes or representative patterns such as centroid.
• Assessment of output – One output gives the data domain or sometimes by a clustering algorithm. The output also comes from cluster validity analysis.

DATA CLUSTERING TECHNIQUES

Data clustering can be done by two methods –

Hierarchical clustering

Partitonal clustering

1. Hierarchical clustering

Hierarchical clustering algorithm includes seven patterns three clusters of the given statistics data. Further it is classified into a dendogram representing the grouping of patterns and similarity levels at which the grouping change. It can be broken at different levels to provide different clustering’s of the data. This algorithm is further divided into-

• Single link – In single link clustering method, the distance between two clusters is the minimum of the distances between all pairs of patterns drawn from two clusters. It produces a chain effect betweenclusters.
• Complete link - In complete link clustering method, the distance between two clusters is maximum of the distances between all pairs of patterns drawn from two clusters. It produces compact and tightly bound clusters.

2. Partitional Clustering

A partitional clustering algorithm obtains a single partition of the data instead of the clusters. A problem which accompanies the use of partitional algorithm is the choice of the number of desired output clusters. The algorithm is further divided into –

• Squared error – It tends to work well with isolated and compact clusters. In this algorithm the ‘k’ means is the simplest form of criterion. It starts with a simple partition and reassigns the patterns to the clusters based on the similarity between the patterns and the cluster centres until a convergence criterion is met.
• Graph – theoretic – It is based on the minimal spanning tree of the data.
• Mixture resolving - The assumption is that the patterns to be clustered are drawn from one of the several distributions and the goal is to identify the parameters of each pattern with their number. The Expectation Maximization algorithm of missing data has been applied to the problem of parameter estimation.