A study on the factors considered when choosing an appropriate data mining algorithm

PRESENTED BY:-
JYOTIR MOY CHATTERJEE

 ABSTRACT
 INTRODUCTION
 BACKGROUND INFORMATION
 LITERATURE REVIEW
 METHODOLOGY
 DISCUSSION
 FUTURE WORK
 CONCLUSION
 REFERENCE

 A lot of data is generated and collected in today’s organizations.
 Businesses now a days analyze the data to make business decisions.
 Various algorithms may be used to analyze the data however some of
them do not yield useful knowledge.
 Choosing the appropriate algorithm remains a problem given the
diversity in available algorithms.
 There are many algorithms, it is difficult for analysts and researchers to
choose algorithm will be suitable for their needs.
 A number of factors to be considered when selecting an algorithm are
discussed to help analysts in choosing appropriate algorithms.

 Data mining is a computer-assisted process of discovering patterns in
data through analyzing huge amounts of data and extracting meaning
from these patterns, [1].
 Businesses are now being forced to use the data they have to their
advantage over their competitors.
 The uncovered information may be used by the organizations for market
segmentation, fraud detection, market basket analysis and marketing.
 In order to extract this useful information, there are many algorithms and
tools that may be used. They may be categorized into classification,
regression, segmentation, association and sequence analysis algorithms,
[2].
 The data mining process follows a set of steps that must be executed
irregardless of the algorithm that will be implemented. Several
algorithms may be used to perform the same task and still produce
different results.

 Most organizations are exposed to a number of data mining algorithms
such that selecting an appropriate algorithm is a difficult task.
 When an inappropriate algorithm is implemented, the discovered
knowledge is often meaningless to the organization as it does not uncover
the correct information leading to wrong business decisions, [7].
 There are no specific guidelines as to how a researcher or analyst may
choose an algorithm .
 As a result, inappropriate algorithms have been used and no useful
information uncovered.

Fig 1: Steps Involved in the Data Mining Process

 It is very important that an appropriate algorithm be used as it has an
impact on the results and knowledge derived.
 The algorithm defines the concrete method/s that will be used to search
for patterns in data.
 In other words, the algorithm used should be able to yield the desired
patterns.
 An organization may use a feature extraction model to create a set of
predictors and then a classification model to make a prediction, [3].
 Researchers and data miners choose the data mining algorithm by
considering the main goal of the problem to be solved and the structure
of the data set, [7]
 Selecting an algorithm has become so complex that other researchers
adopt more than one technique in an effort to improve the data mining
results,[6]
 The algorithm must produce the desired results based on the data mining
task that needs to be accomplished and must be fast enough to search
through the enormous amounts of data.

 Evaluation measures and the configuration parameters are also critical in
choosing a data mining algorithm.
 data miners must select an algorithm that may be fully integrated with the
database or data warehouse so as to avoid cost of extracting, importing and
analyzing the data.
 For instance if the task is a classification task, then the possible algorithms would
be logistic regression, Naïve Bayes, Support vector machine or decision trees.
 When a task is an association task, the Apriori algorithm may be implemented
[4].
 An appropriate algorithm is one that will provide the data miners with the correct
results and useful knowledge that will inform the business decisions
 Research shows that there are no predefined guidelines that may be followed in
selecting a data mining algorithm.
 A careful consideration of the different factors identified by different authors
must help data miners in selecting an algorithm.
 Ideally, a combination of the factors will increase the chances of selecting the
most appropriate data mining algorithm.

 Literature on data mining techniques and algorithms was reviewed to
give an insight into how data miners may select an algorithm to use in
their tasks.
 An analysis of these requirements helped to draw a list of factors that
could be considered when selecting an appropriate data mining
algorithm.
 To understand how the data mining process is carried out, documentation
for two tools , DM assistant, [7] and GESCONDA tool,[8] was reviewed.
 The documentation explained how a user would use the tools for data
mining.
 There were no specific details of how the algorithm could be chosen but
the result sets from the various algorithms were provided.
 It is from these result sets that the factor proved to be important in
selecting an algorithm for use.

 Piatetsky-Shapiro’s Rule-Interest Function:
RI = |X ∩ Y | −|X||Y | ,
N
where N is the total number of tuples, |X| and |Y | are the number of tuples satisfying conditions
X and Y , respectively, |X ∩ Y | is the number of tuples satisfying X → Y , and |X||Y |/N is the
number of tuples expected if X and Y were independent (i.e., not associated).When RI = 0, then X
and Y are statistically independent and the rule is not interesting. When RI > 0(RI < 0), then X is
positively (negatively) correlated to Y .Those rules which do not exceed a pre-determined minimum
significance threshold are determined to be the most interesting.
 Hamilton and Fudger’s I-Measures:
I1 = ∑ c(t(v)),
v
where v is an attribute value, t(v) is the concept hierarchy associated with the attribute containing v,
and c(t(v)) is a function that returns 1 if v is a non-ANY, non-leaf concept, and 0 otherwise. The I2
measure considers the depth and weighted height for all nodes in a summary is given by:
I2 = ∑(k)d(v, t(v)) + (1 − k)wh(v, t(v)),
v
where k specifies the relative significance of the depth of a concept versus the weighted depth (e.g.,
k = 0.5 indicates the distance from the root node is as significant as the distance from the leaf nodes),
v is an attribute value, t(v) is the concept hierarchy associated with the attribute containing v, d(v,
t(v)) is the depth of v in concept hierarchy t(v), and wh(v, t(v)) is the weighted height of v in concept
hierarchy t(v).The depth d(v, t(v)) of v in concept hierarchy t(v) is defined so that the depth of the
root node is zero and the depth of any other concept is one more than the depth of its parent. The
weighted height wh(v, t(v)) of v in concept hierarchy t(v) is a function of the number of leaf concepts
it has as descendants and the sum of the distances to each of its descendants. Summaries associated
with higher values of I1 and I2 are considered more interesting.

 Objective measures are based upon the structure of the discovered
patterns, while subjective measures are based upon user beliefs or
biases regarding relationships in the data.

 This study revealed that there are factors that researchers may consider
when selecting a data mining algorithm.
 Main goal of the problem to be solved
 This factor considers the reason why we are mining the data as well as
the nature of the problem we are trying to address.
 depending on the problem we are trying to solve data miners must select
an appropriate algorithm.

 Structure of the available data set
 The data you provide is first analyzed to identify specific types of
patterns or trends before defining the mining model.
 The results of the analysis are used to define the parameters for the
mining model hence the need to consider the data set,[2].

 Expected results.
 Every data mining task must yield a desired solution hence the
importance of selecting an appropriate algorithm.
 The major aim of data mining is to identify patterns and trends in data so
as to use the knowledge in decision making, [5].

 What the Information Would be used for
 Results of data mining tasks inform the business’s decisions such that if
the uncovered information is wrong, incorrect decisions may be made.
 The information may be used for market segmentation, market basket
analysis and trend analysis to mention a few examples.
 In order for the information to be useful, it must be presented in the
correct structure, for example to decide on the market segments, the
information must be grouped together into clusters thus requiring the
adoption of clustering algorithms for the data mining task.
 Familiarity with an Algorithm
 Having experience in implementing an algorithm may make the selection
much easier.
 This factor however becomes useful when the same type of tasks is to be
performed.

 Configuration Parameters
 When considering this factor, select an algorithm that may be integrated
to the data source at minimal costs.
 An algorithm that may be fully integrated to the organization's database
would be more ideal, where extracting, importing and exporting the data
will be easy to automate without incurring extra costs.

 Taking the proposed factors into consideration may help data miners.
 An appropriate algorithm increases chances of extracting correct, useful
knowledge which gain competitive advantage.
 This stage in the data mining process is very important
 It determines the kind of results that will be generated which will
influence the business decisions.
 Therefore if the results are incorrect, there are chances will be high that
business will make wrong decisions posing a negative impact on its
operations.

 Further work may be conducted to identify more factors that data miners
may consider when choosing a data mining algorithm
 The study is based on literature review and did not experiment to verify
whether the suggested factors really assist in selecting an appropriate
algorithm.

[1]Alexander, D. (n.d.): Data Mining [online]
http://www.laits.utexas.edu/~anorman/BUS.FOR/course.mat/Alex/(acces
sed 13/12/2013 234 p.m.).
[2]Anon (2012): Data Mining Algorithms (Analysis Services – Data
Mining)[online]http://technet.microsoft.com/enus/library/ms175595.aspx
(accessed on 08/01/2014) 2012.
[3] Anon, (2008): Oracle Data Mining Concepts 11g Release 1 (11.1).
[4] Anon, (n. d.): An overview of Data Mining Techniques.
[5]Brown, M. (2012): Data Mining Techniques [online]
http://www.ibm.com/developerworks/library/ba-data mining-techniques/
(accessed 23/12/2013).
[6]Cheeseman, P. and R. W. Oldford (1994): Selecting models from data.
LNStats 89, Springer.
[7] Gibert, K., Sànchez-Marrè, M. and V. Codina (2010): Choosing the Right
Data Mining Technique: Classification of Methods and Intelligent
Recommendation.
[8] Miquel Sànchez-Marrè, Karina Gibert and Ignasi Rodríguez-Roda(n.d.):
GESCONDA: A Tool for Knowledge Discovery and Data Mining in
Environmental Databases.

A study on the factors considered when choosing an appropriate data mining algorithm

More Related Content

What's hot

Viewers also liked

Similar to A study on the factors considered when choosing an appropriate data mining algorithm

Recently uploaded

A study on the factors considered when choosing an appropriate data mining algorithm

Editor's Notes