Cluster analysis is a kind of unsupervised learning, classification is a kind of supervised learning, is to use the known categories of training data to get a classification model, so the main difference between clustering and classification is whether it is necessary to pre-defined categories, that is, cluster analysis relies on the data itself to determine the relationship between the data, So clustering analysis has great advantages, especially for processing large amounts of raw data.
Performance indicators for clustering methods: 1. Scalability, 2, adaptive, 3, robustness 4, explanatory
The data classes used for clustering are: Data matrix dissimilarity matrices
Normalization is done on the basis of the center of the transformation, to ensure that the variable range is equal, the common normalization method has the maximum normalization, the sum of normalization, mean standard deviation normalization and extreme difference normalization
The clustering method includes a method based on the partitioning approach based on the method of density based approach based on grid method
The distance of continuous variable needs to satisfy the conditions of reflexive symmetry and triangular inequality, and Matrix D is a symmetric matrix, and the element on the diagonal is 0.
The similarity coefficient should satisfy the normalization of the reflexive symmetry to satisfy the triangular inequalities
A method of clustering based on segmentation: K-means algorithm K-center value arithmetic
Based on the hierarchical clustering method, the method using distance as a measure of the standard distance of the cluster is: the minimum distance maximum distance mean value, etc.
It is divided into bottom-up and top-down approaches: where bottom-up is a cluster of each object, and by constantly merging these basic clusters to form larger clusters, knowing that the conditions are met
Top-down is the first to see all of the objects are a cluster of objects, and then according to a certain criteria for continuous segmentation of the cluster to form a smaller cluster, thus completing the cluster
Density-based clustering main starvation algorithm has dbscan OPTICS Denclue clique
The main algorithm of clustering based on grid is Sting Wavecluster clique
Model-based clustering mainly includes neural network method and statistical method
Data Mining-Cluster analysis