Database Reference
In-Depth Information
use frequent patterns directly by simply examining them manually. This is different
from many other major data problems such as outlier analysis and classification in
which the output of the process is concise, usually a goal in of itself, and is usually
presented directly to the user for manual inspection. Therefore, this chapter will
focus on applications of frequent pattern mining, which serve as the most important
motivating factor for frequent pattern mining algorithms.
The applications of frequent pattern mining span a very wide variety of fields, and
also incorporate several different data domains. Correspondingly, different kinds of
variations of frequent pattern mining may be used to address the unique problems
which are specific to each domain. For example, the kinds of patterns mined will
very different in the context of temporal, spatial, multimedia or biological data. Some
examples of the wide variety of problem and data domains are as follows:
Customer Analysis:
Customer analysis is the original and motivating applica-
tion for frequent pattern mining. The idea is that frequent correlations between
customer buying behavior can be used in order to make useful business decisions.
Facilitator for other major data mining problems:
Frequent pattern mining
has close connections with other major data mining problems such as clustering
and classification. This is because frequent pattern mining is closely related to
the problem of subspace clustering. Furthermore, discriminative frequent patterns
can often be used to construct classifiers. Since the clustering problem is closely
related to outlier analysis, frequent patterns are often used in order to determine
outliers from the underlying data.
Indexing and Retrieval:
Frequent pattern mining algorithms can be used in order
to design signature-based techniques for indexing and retrieval of market basket
data. Since indexes often depend upon a concise representation of the underlying
data, frequent pattern mining methods serve as an important intermediate step in
the process.
Web Mining Tasks:
Sequential pattern mining algorithms are frequently used to
determine important traversal patterns from Web logs. Such traversal patterns can
be used in order to design and organize Web sites.
Software Bug Detection:
Frequent patterns can be used to determine bugs in
software programs by using frequent pattern mining in order to determine the
most relevant patterns in the underlying data.
Event Detection and Other Temporal Applications:
A variety of temporal ap-
plications such as event detection use frequent pattern mining methods. Many
techniques have been designed for periodic pattern mining, event detection, and
other related applications which use variants of frequent pattern mining methods
as subroutines.
Spatial and Spatiotemporal Analysis:
Spatial data is one in which both spatial
and non-spatial attributes are attached to objects (e.g. temperature readings on the
sea surface). In such cases, association rules can characterize useful relationships
between the spatial and non-spatial properties of the attributes. Spatio-temporal
data such as trajectories can often be analyzed with the use of frequent pattern
