Research of Data Mining Based on Apriori algorithm

  • 格式:pdf
  • 大小:171.62 KB
  • 文档页数:4

Research of Data Mining Based on Apriori algorithm in Cutting Database

Guofeng Wang, Xiu Yu, Dongbiao Peng, Yinhu Cui, Qiming Li School of Mechanical Engineering Tianjin University Tianjin 300072, China

Abstract—Cutting data mining is an important method to increase efficiency㸪discover hidden knowledges in cutting database, and provide guidance for cutting decisions.This paper analyze the Apriori algorithm for association rules mining, and make some improvement for this algorithm based on the features of cutting database. Apriori algorithm is improved to mine association rules in cutting database. The results show that the Apriori algorithm can be efficiently used in cutting data mining, and improved algorithm can achieve expected effect better than traditional algorithm.

Keywords-data mining; association rules mining; cutting database; Apriori algorithm

I. INTRODUCTION Cutting database appHared after the combination of machining technique and database technique which can provide effective data information support for machining process. we can get higher machining efficiency and economic efficiency if the reasonable and optimized cutting data provided by cutting database is used [1].However, it is a pity that it can’t derive enough more useful information so as to provide guidance for actual machining although large amount of cutting parameters, processing knowledges, empirical formulas have been stored. In this sense, it can only be called static cutting database, which result in “explosive data, however lack of knowledge”. So it is important to adopt data mining technique for cutting database and discover hidden knowledges from this cutting database. In this paper, on studying the properties of cutting database, Apriori algorithm is improved before using into cutting data mining and the final results shows that some useful knowledge can be get from these large amount of cutting data.

II.ASSOCIATION RULES MINING AND APRIORI ALGORITHM

Association rules mining is one of the most important research methods in data mining which can obtain some useful knowledge to describe the association between different valuable data items out of a great amount of datas. A lot of algorithms about mining association rules have been presented recently, among which Apriori algorithm is one of typical algorithm. It was introduced in 1993 by Agrawal as a powerful method used to find regularities in data trend [2]. Apriori algorithm is a width-first search arithmetic in which recursive method was adopted. A basic property of “every subset of a frequent item sets is still frequent item set, and every superset of a non-frequent item set is not a frequent item set [3]” has

been used in Apriori algorithm to discover all of the frequent item sets. Details and formulas are given as follows:

C1={candidate 1-itemsets};

L1={cęC1/c.count>=minimum support};

For (k=2; Lk-1󰂏ij; k++) {

Ck=sc_candidate(Lk-1);

For each transaction tęD {

Ct=count_support(Ck,t);

For all candidates cęCt {c.count++}

} Lk={cęCk/c.count>=minimum support};

} Return L=UkLk;

The notations and definitions are given as follows: 1) D=set of transactions; each transaction t is included in D; 2) Lk= set of large k-item sets(set of items having minimum

support);

3) Ck= set of candidate k-item sets(items to be counted);

4) Function of sc_candidate is aimed at using Lk-1 to get Ck;

5) Function of count_support is aimed at using Ck and t to get

all of the candidate k-itemsets that included in t.

The Apriori algorithm finds only the frequent item sets. In order to find the associations rules in the database , we must apply the following improved algorithm [4],that is:

For each frequent item set Lk {

Generate all non-empty subsets of Lk;

For each non-empty subset Ls of Lk {

Output-rule: LsĺLk if

support(Lk)/support(Ls)>=min-confidence}

} The Above steps can be concluded to the following two points: (1) The support of every frequent item sets should be larger than the minimum support threshold, and (2) The

This paper is sponsored by Chinese National Science Fund (50805100) and Chinese National Science & Technology Supporting Program (2008BAF32B11) 978-1-4244-7739-5/10/$26.00 ©2010 IEEE