An ID3-improved Approach of for Optimum Rule Mining Through Granular Computing Search

Hadis Smadi Alinia1,Mahmoud Reza Delavar2, Yiyu Yao3
1.Department of Surveying and Geomatics Eng, University of Tehran, Tehran, Iran
2.Deparment of Surveying and Geomatics Eng Center of Excellence in Geomatics Eng and Disaster Management
3.Department of Computer Science, University of Regina, Sasktchewan, Canada
1.alinia@ut.ac.ir; 2. medelavar@ut.ac.ir; 3. yyao@cs.uregina.ca

Abstract: Rule induction is an area of machine learning in which formal rules are extracted from a set of observations or training dataset. Inducted rules can be expressed as a final result of the decision tree in which each branch represents a possible scenario of decision and its outcome. Existing decision learning algorithms like Iterative Dichotomiser (ID3) is an attribute centered method which may introduce unnecessary attributes in the classification rules. To overcome the problem, coverage and confidence measures are applied to select the most promising attribute-value at each step. The proposed approach is granule centered in which, instead of focusing on the selection of a suitable partition, i.e., a family of granules is defined, a teach step, by values of an attribute. This paper is concentrated on the selection of a single granule. The decision tree learning algorithm ID3 and granular network are successfully applied for information table of test dataset of seismic vulnerability of urban areas in Tehran, capital of Iran.

Keywords: ID3 decision tree; Granule network; Uncertainty; Consistent classification; Seismic vulnerability

AttachmentSize
AliniaAccuracy2010.pdf483.47 KB