MIS372 Lecture Notes - Lecture 8: Hierarchical Clustering, Railways Act 1921, Ground Truth
1 Oct 2018
School
Department
Course
Professor
Document Summary
Clusteing: grouping data instances based on their similarities (selected attributes, data instances: Within a group are highly similar to each other. Across groups are very dissimilar to each other: unsupervised/descriptive analysis with no pre- assumptions. Clustering types: exclusive/strict partitions (hard clustering, overlapping partitions (soft clustering, hierarchical partitions (hierarchical clustering, fuzzy/probabilistic partitions (fuzzy clustering) Which clusters each data instance into every group with a specific probability. Clusters are formed based on similarity measures. Centres are vectors (including several selected attributes) Centres may not be part of the original data set. Clusters are regions in the data with high densities. Clusters are separated with low-density data areas. Some data may be lost (in low-density areas) The similarities of data instances are used to find hierarchies of clusters. The output is a dendrogram (tree diagram) Top-down: start with all data in one cluster, separate data points in several iterations, until individual data instance are reached.
