英语论文网

MFILM: a multi-dimensional fuzzy inductive learning method
YAO-TSUNG CHEN*y and BINGCHIANG JENGz
yInstitute of Information Science, Academia Sinica, Taiwan
代写英文论文zDepartment of Information Management,
National Sun Yat-sen University, Taiwan
(Received 8 March 2005; revised 20 June 2005; accepted 22 June 2005)
Inductive learning that creates a decision tree from a set of existing
examples is shown to be useful for automated knowledge acquisition.
Most of the existing methods however, handle only single-dimensional
decision problems. Only some methods can deal with multi-dimensional
decision problems. However, they are based on crisp concepts that are
weak in handling marginal cases. In this paper, we present a multidimensional
fuzzy inductive learning method that integrates the fuzzy
set theory into the conventional multi-dimensional decision tree induction
methods. The method converts a multi-dimensional decision tree
into a fuzzy multi-dimensional decision tree in which hurdle values for
splitting branches and classes associated with leaves are fuzzy. Results
from empirical tests indicate that the new fuzzy approach outperforms
the other conventional methods.
Keywords: Inductive learning; Machine learning; Expert systems;
Multi-dimensional decision tree
1. Introduction
Inductive learning that generates decision trees or decision rules from existing cases
is an important approach for automated acquisition of expert knowledge. Applications
have been reported in many areas such as stock prediction (Braun and
Chandler 1987), credit card application (Carter and Catlett 1987), graduate
admission (Chung and Silver 1992), inventory accounting method choice (Liang
et al. 1992), loan evaluation (Messier and Hansen 1988, Shaw and Gentry 1988) and
medical diagnosis (Michalski et al. 1986, Liang et al. 1992). Their findings indicate
that knowledge induced from these methods is equally or more accurate than
statistical discriminant analysis (Fisher 1936) or other competing models in
predicting new cases.
*Corresponding author. Email: ytchen@mail.iis.sinica.edu.tw
Journal of Experimental & Theoretical Artificial Intelligence,
Vol. 17, No. 3, September 2005, 267–281
Journal of Experimental & Theoretical Artificial Intelligence
ISSN 0952–813X print/ISSN 1362–3079 online # 2005 Taylor & Francis
https://www.tandf.co.uk/journals
DOI: 10.1080/09528130500281828
A typical inductive learning process includes three steps. First, each attribute
domain is partitioned into segments so that boundaries differentiating classes can be
determined. This step determines the hurdle values for different classes. In credit card
analysis, for instance, we may find that the salary of credit card holders can be
partitioned into two segments at US$30 000. That is, if the salary of an applicant
is greater than US$30 000, then its credit is good. If the salary is less than or equal
to US$30 000, then its credit classification is bad.
Following the segmentation of attributes, the discriminant power of each attribute
is analysed. In a classic method called ID3 (Iterative Dichotomizer 3) (Quinlan
1979), the partition and discriminant power of an attribute are determined by
a measurement called entropy gain. The attribute with a higher entropy gain is
considered having a higher discriminant power. Finally, a decision tree 本论文由英语论文网提供整理，提供论文代写，英语论文代写，代写论文，代写英语论文，代写留学生论文，代写英文论文，留学生论文代写相关核心关键词搜索。