A Method of Classification and Association Rules Obtaining

摘要

Data mining popularly known as data or knowledge discovery is the process ofanalyzing data from different perspectives and summarizing it into useful information-information that can be used to increase revenue, cuts costs, or both. Being an interestingfield, data mining has attracted many researchers. Extensive researches have beenconducted on major data mining techniques but few researches have addressed theintegration of these techniques.
　　 My thesis focuses on how to integrate the two major data mining techniques namely,classification and association rules to come up with An Optimal Class Association RuleAlgorithm (OCARA). It is proposed by our study group. Classification and associationrule mining algorithms are two important aspects of data mining. Classificationassociation rule mining algorithm is a promising approach for it involves the use ofassociation rule mining algorithm to discover classification rules.
　　 OCARA inherits the strength of Classification and association rule mining algorithms.Because of this reason, OCARA is a powerful algorithm when compared to eitherClassification or Association rule mining algorithms.
　　 To.verify the strength of OCARA, we conducted experiment using eight different datasets of UCI (University of California at Irvine). We compared OCARA with other threepopular algorithms (C4.5, CBA, RMR). The end result proved that the support thresholdwas greatly influenced by the rule accuracy and the rule number. If the support thresholdis between 0.02 and 0.03, the accuracy will be much better, as discussed in this paper.The support threshold was set as 0.02, and the confidence was set as 0.30 in our work.Therefore, OCARA proved to more efficient when compared with others making it morerobust in terms of its accuracy.
　　 The reason for OCARA's high accuracy is because of optimal association rule miningalgorithm and the rule set is sorted by priority of rules resulting into a more accurateclassifier. Therefore, we can confidently say OCARA is an accurate classifier and hasbetter performance and is more efficient when compared with C4.5, CBA, and RMRalgorithm.
　　 This thesis makes major contribution to this young industry of data mining since it hassuccessfully proposed and tested a new algorithm, OCARA.
　　 However, OCARA has many rules when compared with RMR when the support islower. To overcome this limitation of having many rules, we are encouraging othersresearchers to focus on this promising algorithm by improving its efficiency.

目录

英文文摘

DEDICATION

LIST OF FIGURES

LIST OF TABLES

CHAPER Ⅰ:INTRODUCTION

1.1 General Introduction

1.2 Data Mining in General

1.3 Benefit ofData Mining

1.3.1. Advantages of Data Mining

1.3.2. Disadvantages of Data Mining

1.4 Data Mining Applications

1.5 Data Mining Techniques

1.6 Data Mining Process

1.7 Motivation and Contribution of Thesis

1.8 Outline

CHAPTER Ⅱ:LITERATURE REVIEW

2.1 Types of Rules Related to this Thesis

2.1.1 What is a Rule?

2.1.2 What to do with a Rule?

2.1.3 Caveat:Rules do not imply Causality

2.1.4 How to Evaluate the Rule

2.2 Classification Rules

2.3 Association Rule

2.3.1 Finding Large Itemsets

2.3.2 Generating Rules

2.4 Optimal Association Rule Set

2.4.1 Class Association Rules

2.4.2 Optimal Rule Set

2.5 Types of Algorithms Used for Comparison in this Thesis

2.5.1 C4.5 Algorithm

2.5.2 CBA Algorithm

2.5.3 RMR Algorithm

CHAPTER Ⅲ:BASIC CONCEPTS AND THEORY

3.1 Introduction

CHAPTER Ⅳ:OCARA Algorithm

4.1 Introduction

4.2 Discovering the Optimal Rules Set

4.3 Sorting Rules

4.4 Matching Rules

CHAPTER Ⅴ:EXPERIMENTAL RESULTS

5.1 Materials

5.2 Description of the Environment

5.2.1 Hardware

5.2.2 Software

5.3 Results and Data Analysis

5.3.1 Graph of the prediction accuracy of C4.5 and OCARA algorithms

5.3.2 Graph of the prediction accuracy of CBA and OCARA algorithms

5.3.3 Graph of the prediction accuracy of RMR and OCARA algorithms

5.3.4 Graph of the prediction accuracy of C4.5,CBA and OCARA algorithms

5.3.5 Graph of the prediction accuracy of C4.5,RMR and OCARA algorithms

5.5.6 Graph of the prediction accuracy of CBA,RMR and OCARA algorithms

5.3.7 Graph of the prediction accuracy of C4.5,CBA,RMR and OCARA algorithms

CONCLUSION

REFERENCES

ACKNOWLEDGMENT

附录