Introduction - Data Preprocessing in Data Mining - page 8

1.4.1 Pattern Mining [ 25 ]
It is adopted as amore general termthan frequent patternmining or associationmining
since pattern mining also covers rare and negative patterns as well. For example, in
pattern mining, the search of rules is also focused on multilevel, multidimensional,
approximate, uncertain, compressed, rare/negative and high-dimensional patterns.
The mining methods do not only involve candidate generation and growth, but also
interestingness, correlation and exception rules, distributed and incremental mining,
etc.
1.4.2 Outlier Detection [ 9 ]
Also known as anomaly detection, it is the process of finding data examples with
behaviours that are very different from the expectation. Such examples are called
outliers or anomalies. It has a high relation with clustering analysis, because the latter
finds the majority patterns in a data set and organizes the data accordingly, whereas
outlier detection attempts to catch those exceptional cases that present significant
deviations from the majority patterns.
1.5 Other Learning Paradigms
Some DM problems are being clearly differentiated from the classical ones and some
of them even cannot be placed into one of the two mentioned learning categories,
neither supervised or unsupervised learning. As a result, this section will supply a
brief description of other major learning paradigms which are widespread and recent
challenges in the DM research community.
We establish a general division based on the nature of the learning paradigm.
When the paradigm presents extensions on data acquirement or distribution, imposed
restrictions on models or the implication of more complex procedures to obtain
suitable knowledge, we refer to extended paradigm. On the other hand, when the
paradigm can only be understood as an mixture of supervised and unsupervised
learning, we refer to hybrid paradigm. Note that we only mention some learning
paradigms out of the universe of possibilities and its interpretations, assuming that
this section is just intended to introduce the issue.
1.5.1 Imbalanced Learning [ 22 ]
It is an extended supervised learning paradigm, a classification problem where the
data has exceptional distribution on the target attribute. This issue occurs when the