Oracle® Data Mining Concepts 11g Release 2 (11.2) E1680807 


PDF · Mobi · ePub 
This chapter describes the enhanced kMeans clustering algorithm supported by Oracle Data Mining.
See Also:
Chapter 7, "Clustering"This chapter includes the following topics:
The kMeans algorithm is a distancebased clustering algorithm that partitions the data into a specified number of clusters.
Distancebased algorithms rely on a distance function to measure the similarity between cases. Cases are assigned to the nearest cluster according to the distance function used.
Oracle Data Mining implements an enhanced version of the kMeans algorithm with the following features:
Distance function — The algorithm supports Euclidean, Cosine, and Fast Cosine distance functions. The default is Euclidean.
Hierarchical model build —The algorithm builds a model in a topdown hierarchical manner, using binary splits and refinement of all nodes at the end. In this sense, the algorithm is similar to the bisecting kMeans algorithm. The centroids of the inner nodes in the hierarchy are updated to reflect changes as the tree evolves. The whole tree is returned.
Tree growth — The algorithm uses a specified split criterion to grow the tree one node at a time until a specified maximum number of clusters is reached, or until the number of distinct cases is reached. The split criterion may be the variance or the cluster size. By default the split criterion is the variance.
Cluster properties — For each cluster, the algorithm returns the centroid, a histogram for each attribute, and a rule describing the hyperbox that encloses the majority of the data assigned to the cluster. The centroid reports the mode for categorical attributes and the mean and variance for numerical attributes.
This approach to kMeans avoids the need for building multiple kMeans models and provides clustering results that are consistently superior to the traditional kMeans.
The centroid represents the most typical case in a cluster. For example, in a data set of customer ages and incomes, the centroid of each cluster would be a customer of average age and average income in that cluster. The centroid is a prototype. It does not necessarily describe any given case assigned to the cluster.
The attribute values for the centroid are the mean of the numerical attributes and the mode of the categorical attributes.
The Oracle Data Mining enhanced kMeans algorithm supports several buildtime settings. All the settings have default values. There is no reason to override the defaults unless you want to influence the behavior of the algorithm in some specific way.
You can configure kMeans by specifying any of the following:
Number of clusters
Growth factor for memory allocated to hold clusters
Convergence tolerance
Distance Function. The default distance function is Euclidean.
Split criterion. The default criterion is the variance.
Number of iterations for building the cluster tree.
The fraction of attribute values that must be nonnull in order for an attribute to be included in the rule description for a cluster. Setting the parameter value too high in data with missing values can result in very short or even empty rules.
Number of histogram bins. The bin boundaries for each attribute are computed globally on the entire training data set. The binning method is equiwidth. All attributes have the same number of bins with the exception of attributes with a single value that have only one bin.
See Also:
Oracle Database PL/SQL Packages and Types Reference for details about the build settings for kMeansNormalization is typically required by the kMeans algorithm. Automatic Data Preparation performs outliersensitive normalization for kMeans. If you do not use ADP, you should normalize numeric attributes before creating or applying the model.
When there are missing values in columns with simple data types (not nested), kMeans interprets them as missing at random. The algorithm replaces missing categorical values with the mode and missing numerical values with the mean.
When there are missing values in nested columns, kMeans interprets them as sparse. The algorithm replaces sparse numerical data with zeros and sparse categorical data with zero vectors.
Oracle Database PL/SQL Packages and Types Reference for details about normalization routines
Chapter 19 for information about automatic and embedded data transformation in Oracle Data Mining
Oracle Data Mining Application Developer's Guide for information about support for nested columns and missing data in Oracle Data Mining