Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!nntp.club.cc.cmu.edu!hudson.lm.com!news.pop.psu.edu!news.cac.psu.edu!howland.reston.ans.net!gatech!newsxfer.itd.umich.edu!zip.eecs.umich.edu!caen!night.primate.wisc.edu!hpg30a.csc.cuhk.hk!hkuxb.hku.hk!hkusua!yhfung
From: yhfung@hkusua (Fung Yun Hoi)
Subject: Re: Multicollinearity and outliers
Message-ID: <D0p57v.B9u@hkuxb.hku.hk>
Sender: usenet@hkuxb.hku.hk (USENET News System)
Nntp-Posting-Host: hkusua.hku.hk
Organization: The University of Hong Kong
X-Newsreader: TIN [version 1.1 PL6]
References: <3cdmm6$70c@celebrian.otago.ac.nz>
Date: Mon, 12 Dec 1994 11:57:31 GMT
Lines: 34

argray@rivendell.otago.ac.nz wrote:
: Hello everyone,
: 	I recently posted to this group regarding the effect of
: multicollinearity in training data on neural networks, specifically
: backpropagation trained networks.  The only email replies I have received were
: from people who would also like information and references on this subject. 
: Since there does seem to be some interest in this could anyone who does know 
: anything about it please post to the group.  I would also be interested in the
: effects of multiple outliers in the dataset on the trained network.
: 	ANY information or references would be MOST appreciated.  Thanks in
: advance (I hope).
: 
: Andrew Gray
: 
I have come across these matters when I was doing my MSc distertation
on using artificial neural networks (ANNs) for forecasting electricity 
consumption in Hong Kong. Since my supervisor, Dr. V.M. Rao Tummala, is
very familiar with the classical forecasting methods especially the
multiple linear regression in which a lot of assumptions have to be
made and validated by the residual plots. Outliers should be removed
manually before feeding the data to the inputs of ANNs. The activation
function for each neurons that I used is linear therefore the ANNs
made can have good generalisation property.  

In fact, you may have the literature on multicollinearity from the
classical forecasting books.  

Regards,

Y.H. Fung
yhfung@hkueee.hku.hk
Dept. of Electrical & Electronic Engineering
The University of Hong Kong

