Depicting diversity in rules extracted from ensembles

Chan, Fabian H. P. and Ali Chekima, and Sitiol, Augustina and Kalaiarasi Sonai Muthu, Anbananthen (2009) Depicting diversity in rules extracted from ensembles. In: 6th International Symposium on Neural Networks, 2009, Wuhan, China.

Full text not available from this repository.


Ensembles are committees of neural networks used to achieve better classification and generalization accuracy in contrast to just using a single neural network. Traditionally incomprehensible models such as artificial neural networks presents a problem in its application to safety critical domains where not only accuracy but also model transparency is a requirement. This problem is not only inherited but further multiplied in ensembles. Furthermore, the aspect of diversity by which ensembles achieve improved classification ability cannot be reflected in rule extraction methods designed for single neural networks hence the need for rule extraction methods specifically designed for ensembles. This paper presents a decompositional rule extraction algorithm for ensembles which is able to approximately decompose the neural networks in an ensemble and reflect their collective diversity to identify significantly contributing inputs.

Item Type: Conference or Workshop Item (UNSPECIFIED)
Uncontrolled Keywords: Safety critical domain, Ensemble neural network, Rule extraction, Comprehensibility, Data mining
Subjects: Q Science > QA Mathematics > QA76 Computer software
Divisions: SCHOOL > School of Engineering and Information Technology
Depositing User: Unnamed user with email
Date Deposited: 18 Aug 2011 08:14
Last Modified: 30 Dec 2014 06:19

Actions (login required)

View Item View Item