Evaluating Learning Algorithms Composed by a Constructive Meta-learning Scheme for a Rule Evaluation Support Method - Mining Complex Data - page 89

6

Evaluating Learning Algorithms Composed by a
Constructive Meta-learning Scheme for a Rule
Evaluation Support Method

Hidenao Abe 1 ,ShusakuTsumoto 1 ,MihoOhsaki 2 , and Takahira Yamaguchi 3

1

Department of Medical Informatics, Shimane University, School of Medicine
[e-mail], [e-mail]

2

Faculty of Engineering, Doshisha University
[e-mail]

3

Faculty of Science and Technology, Keio University
[e-mail]

Abstract. The Post-processing of mined patterns such as rules, trees and so forth is
one of the key operations in a data mining process. However, it is di cult for human
experts to completely evaluate several thousand rules from a large dataset with noise.
To reduce the cost of this kind of rule evaluation task, we have developed a rule eval-
uation support method with rule evaluation models that learn from objective indices
for mined classification rules and evaluations by a human expert for each rule.
In this paper, we present an evaluation of the learning algorithms of our rule eval-
uation support method for the post-processing of mined patterns with rule evaluation
models based on objective indices. To enhance the adaptability of these rule evalua-
tion models, we introduced a constructive meta-learning system for the construction of
appropriate learning algorithms. We then have performed case studies using the menin-
gitis as an actual problem. Furthermore, we evaluated our method with the eight rule
sets obtained from eight UCI datasets. With regard to these results, we show the ap-
plicability of the constrictive meta-learning scheme as a learning algorithm selection
method for our rule evaluation support method.

6.1 Introduction

In recent years, enormous amounts of data have been stored on information
systems in natural science, social science, and business domains. People have
been able to obtain valuable knowledge due to the development of information
technology. In addition, data mining techniques combine different kinds of tech-
nologies such as database technologies, statistical methods, and machine learning
methods. Data mining has become well known as a method for utilizing the data
stored on database systems. In particular, if-then rules, which are produced by
rule induction algorithms, are considered to be a highly usable and readable
output of data mining. However, for large datasets with hundreds of attributes
including noise, the process often obtains many thousands of rules. From such
a large rule set, it is di cult for human experts to find out valuable knowledge,
which is rarely included in the rule set.