Please use this identifier to cite or link to this item: http://scholarbank.nus.edu.sg/handle/10635/27770
Title: An incremental approach to contribution - based feature selection
Authors: LIU JUN
Keywords: Incremental training, Feature selection, Neural networks, Feedforward neural network, Supervised learning, Input attributes.
Issue Date: 23-Oct-2003
Source: LIU JUN (2003-10-23). An incremental approach to contribution - based feature selection. ScholarBank@NUS Repository.
Abstract: In this thesis, a novel incremental approach for Neural Network (NN) feature selection is presented. Two related topics a?? NN training and feature selection - are investigated. Namely ITID, the new incremental training approach works by dividing the whole input dimension into several sub-dimensions each of which corresponds to an input attribute. Instead of learning input attributes altogether as an input vector in a training instance, NN learns input attributes one after another through their corresponding sub-networks and the NN structure is grown incrementally with an increasing input dimension. Two novel feature selection methods based on the aforementioned incremental neural network training approach are further investigated. Using error rate as the evaluation function, feature selection is incorporated with incremental training. If an incoming attribute is consistent with the previous attributes and relevant to the output target, the network performance will be improved, otherwise the network performance will be degraded. Through a network performance snapshot, NN can directly find out the attributes that make no/little contribution. These attributes are marked and considered for removal.
URI: http://scholarbank.nus.edu.sg/handle/10635/27770
Appears in Collections:Master's Theses (Restricted)

Show full item record
Files in This Item:
File Description SizeFormatAccess SettingsVersion 
LiuJun.pdf347.8 kBAdobe PDF

RESTRICTED

NoneLog In

Page view(s)

263
checked on Dec 18, 2017

Download(s)

3
checked on Dec 18, 2017

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.