Data Mining: Practical Machine Learning Tools and Techniques, Second Edition

SimpleKMeansclusters data using k-means; the number of clusters is specified

by a parameter.Cobwebimplements both the Cobweb algorithm for nominal

attributes and the Classit algorithm for numeric attributes. The ordering and

priority of the merging and splitting operators differs between the original

Cobweb and Classit papers (where it is somewhat ambiguous). This imple-

mentation always compares four different ways of treating a new instance and

chooses the best: adding it to the best host, making it into a new leaf, merging

the two best hosts and adding it to the merged node, and splitting the best host

and adding it to one of the splits.Acuityand cutoffare parameters.

FarthestFirstimplements the farthest-first traversal algorithm of Hochbaum

and Shmoys (1985), cited by Sanjoy Dasgupta (2002); a fast, simple, approxi-

mate clusterer modeled on k-means.MakeDensityBasedClustereris a meta-

clusterer that wraps a clustering algorithm to make it return a probability

distribution and density. To each cluster it fits a discrete distribution or a

symmetric normal distribution (whose minimum standard deviation is a

parameter).

10.7 Association-rule learners

Weka has three association-rule learners, listed in Table 10.8.Aprioriimple-

ments the Apriori algorithm (Section 4.5). It starts with a minimum support of

100% of the data items and decreases this in steps of 5% until there are at least

10 rules with the required minimum confidence of 0.9 or until the support has

10.7 ASSOCIATION-RULE LEARNERS 419

Table 10.7 Clustering algorithms.

Name Function

EM Cluster using expectation maximization

Cobweb Implements the Cobweb and Classit clustering algorithms

FarthestFirst Cluster using the farthest first traversal algorithm

MakeDensityBasedClusterer Wrap a clusterer to make it return distribution and density

SimpleKMeans Cluster using the k-means method

Table 10.8 Association-rule learners.

Name Function

Apriori Find association rules using the Apriori algorithm

PredictiveApriori Find association rules sorted by predictive accuracy

Tertius Confirmation-guided discovery of association or classification rules