Building a highly-compact and accurate associative classifier |
| |
Authors: | Xing Zhang Guoqing Chen Qiang Wei |
| |
Affiliation: | 1.Department of Management Science and Engineering, School of Economics and Management,Tsinghua University,Beijing,China |
| |
Abstract: | Associative classification has aroused significant research attention in recent years due to its advantage in rule forms with
satisfactory accuracy. However, the rules in associative classifiers derived from typical association rule mining (e.g., Apriori-type)
may easily become too many to be understood and even be sometimes redundant or conflicting. To deal with these issues of concern,
a recently proposed approach (i.e., GARC) appears to be superior to other existing approaches (e.g., C4.5-type, NN, SVM, CBA)
in two respects: one is its classification accuracy that is equally satisfactory; the other is the compactness that the generated
classifier is constituted with much fewer rules. Along with this line of methodological thinking, this paper presents a novel
GARC-type approach, namely GEAR, to build an associative classifier with three distinctive and desirable features. First,
the rules in the GEAR classifier are more intuitively appealing; second, the GEAR classification accuracy is improved or at
least as good as others; and third, the GEAR classifier is significantly more compact in size. In doing so, a number of notions
including rule redundancy and compact set are provided, together with related properties that could be incorporated into the
rule mining process as algorithmic pruning strategies. The experimental results with benchmarking datasets also reveal that
GEAR outperforms GARC and other approaches in an effective manner. |
| |
Keywords: | |
本文献已被 SpringerLink 等数据库收录! |
|