A New Multiobjective Evolutionary Algorithm

for Mining a Reduced Set of Interesting Positive

and Negative Quantitative Association Rules
Abstract:

Many algorithms are used for quantitative likes rules that mainly focused on optimistic
dependences but these algorithms most often ignore the negative dependencies. Hence the
algorithms are used for extracting useful information and measure the quality of evaluation
criteria. Currently, researchers work on some algorithm of association rules for the
multiobjective issues by using the datasets. In this research article, the research presented the
MOPNA (a new multiobjective evolutionary algorithm) for mining the negative and positive
quantitative association rule in less cost of computations. This algorithm also performs the
attributes if evolutionary learning and the situation for every rule selection. While research also
focusing to introduce the external population and to restart a process for storing the
nondominated rules for the set of rules improvement. This research work is focusing major 3
points one is to discuss the compensability, interestingness, and the performances for obtaining
the rules that will further easy for understanding and giving the best data sets.

Keywords: Data Mining, Multiobjective Evolutionary Algorithm, Quantitative Association

Rule.

Introduction:

In the previous year, the digital revolution gives accessible methods for storing and collected
data. For the huge number of data and infinite increase of data for extracting useful information
is even more complicated as it can cause the problem of complexity and scalability. Hence the
data mining techniques are introduced that widely used more commonly in each domain for
extracting the presentable information for giving a large number of datasets. However, some
association rules also identify that present the dependencies in terms of datasets items. Moreover,
the previous research also focused on the association of mining rules that analyzes the discrete as
well as binary rules. research demonstrates that data is consists of quantitative values therefore
multiple studies have been analyzed for mining quantitative association rules for measuring the
quantitative values of data sets.
In this research work, the author observed that many algorithms focused on the positive QAR
without focusing on the negative QARs, therefore, researchers tried to analyze the method for
both negative and positive association rule of mining in the given mining data. The research
objective is to analyze the negative association rule mining and also discuss the itemset and their
working as well as demonstrate how negative association rules can be more effective in
measuring the quantitative data.
In this research work, the genetic algorithm is considered t be the most powerful research
technique for extracting the set of QARs from the given datasets. These algorithms used to
evaluate and measure the quality of the association rule. Hence specific association rules for the
multiobjective issues are identified for extracting the association rules. Therefore, this research
permits the research to focused on the rules for mining and measuring the set of rules regularly.
However, the multiobjective evolutionary algorithm (MOEAs) presented in this research work as
it gives valid solutions that meet the presented criteria. Hence, this research algorithm is applied
to mine the QARs for measuring the different objectives. MOEAs is purely relying on
decomposition that decomposes the different optimization problem in less cost of computation
also provide the least complexity.
In this research work a novel form of MOEA, MOPNAR are presented that gives less cost in
terms of computation and also reduced the complexity of positive and negative QARs. This
method is easy to understand and best to tradeoff between the support, coverage, and rules of the
datasets. However, this research extends the method of decomposition for performing the
evaluation and selection learning in the attributed to every association rule.
For analyzing the algorithm performances, the algorithm also presented the experimental results
in a variable with the maximum range of 4-91, and also the sample range is between 40 to 22784.
Hence this research work on
 To analyze the performances of presented methods with the comparison of other
evolutionary techniques of mining.
 Comparatively analysis is performed by comparing with other presented approaches such
as MGA, GAR, ARMMGA.
  Results are analyzed and compared with the classical approach of association rules and
classical MOEA.
 Moreover, the research analysis is performed for obtaining the rules of the association.

What is Positive and Negative Quantitative Association Rules?

Previous research observed on the mining association rule based on binary and discrete values,
hence, in the real world. The data is mostly consisting of quantitative values; therefore, the
association rules can be found out by the values of quantitative that is famous in the name of
QARs. The QARs used in many real-world domains mostly in data science in the field of health
and biology. it is difficult to use directly because of the numerical attribute’s values. However,
the negative association rule is also be considered to be the most effective for extracting useful
information also it supports the application’s decisions as well. The positive association rule
contains positive items while the negative association rule contains at least one negative item.
Both rules are used to measure with the QARs that support the dataset items.

New MultiObjective Evolutionary Algorithm for Mining Positive and Negative

Quantitative Association Rules: (MOPNAR)

The proposed algorithm is used to decrease the set of PNQARs with the best trade-off in terms of
rules and for supporting and coverage the presented objectives. For performing the evolutionary
learning mode two other models are introduced such as EP and Restarting process.
EP and Restarting Process within MOEA/D-DE:
This process decomposes the multiobjective optimization issues in Scaler optimization and
further used them to optimize further subproblem. For storing the nondominated association
rules and introducing the multiple diversities of the population and improving the datasets
diversities and presented an EP and Restarting process for the evolutionally model of MOEA.
And EP rules presented and updates the rules for the population solutions. However, the non-
dominated eliminate the EP for avoiding the rules of overlapping. However, EP is consisting of
limited sets of rules because it containing the non-dominance criteria that permit to maintain the
rules.
Objective:
This research is focused on major three objectives for solving the problems of large volume
datasets. Interestingness, Comprehensibility, and performances.
Performance is a measure for improving the coverage of huge datasets that can extract useful
information and knowledge from the datasets. Performance is also enhancing the support and
CF.
The research is focused to find out the strong rules that highly support the itemset. Some
negative association rules are identified that permits to present the negative dependences,
Such as the following rules must be satisfying:
1) CF (X → Y) > 0;
2) support (X → Y) > minSup;
3) ￢ (support (X → Y) > (1 - minSup))

This MinSup is the minimum coverage that used to fulfil that dataset rules.

Interestingness: to find out how interestingly we can measure the rule for permitting them to
extract the useful information according to the user interest. This measure can possibly analyze
the negative association and dependencies and the positive association and dependencies
between the items and ranges.
Comprehensively: this allows us to quantify the rules for understanding. However, rules are
generated that can contain a large number of attributes.

Conclusion:
in this research work MOPNAR and new MOEA used for mining and decreasing the set of
PNQARs. The PNQARs methods are easy to understand, interesting the give the best coverage
for the datasets, that used to achieve the best performances, comprehensibility, and
performances.
This research is focused on the MOEA MOEA/D-DE, for performing the evolutionary learning
for the attributes and rules for the selection conditions. Moreover, this research also introduced
the EP and restarting process for the model of evolutionary for storing all the rules of
nondominated for improving the diversities. It also presented the high dependencies in the items
and solve the drawback of support for the major issues. However, the results are obtained by
using the huge set of datasets also PNQAR gives the best trade-off for the best support, coverage,
and rules. Furthermore. MOPNAR gives the less eliminated sets by giving the less attributed
and easy to understand. In the last the presented method gives the best computational cost for all
the huge datasets and best scalability for all sizes of the problems.