Newsgroups: comp.ai.neural-nets
From: jimmy@ecowar.demon.co.uk (Jimmy Shadbolt)
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!gatech!howland.reston.ans.net!news.sprintlink.net!demon!ecowar.demon.co.uk!jimmy
Subject: Re: Multicollinearity in input data 
Distribution: world
References: <1994Nov24.180600.1@otago.ac.nz>
Organization: Econostat
Reply-To: jimmy@ecowar.demon.co.uk
X-Newsreader: Simple NEWS 1.90 (ka9q DIS 1.21)
Lines: 30
Date: Thu, 24 Nov 1994 15:30:53 +0000
Message-ID: <785691053snz@ecowar.demon.co.uk>
Sender: usenet@demon.co.uk

In article <1994Nov24.180600.1@otago.ac.nz> argray@otago.ac.nz writes:

>Hello everyone,
>
>        I am currently using a back propagation neural network to estimate 
>software effort based on complexity indicators.  I am interested in the 
>effects of some of the input data being multicollinear.  The existence of 
>multicollinearity can be identified using standard statistical techniques and 
>common sense about the nature of the data.  Is it better if I only use 
>uncorrelated inputs or can the network handle the correlations?
>        I have not been able to find any references to this problem and would
>appreciate any assistance in finding references or advice in solving the
>problem.
>        Please reply by email and if there is sufficient interest then I will 
>post a summary to the group.
>
>Thanks in advance,
>Andrew
>

        Many econometricians are facing this problem - you can detect
        and remove collinearity in purely linear context. But what is
        "nonlinear" collinearity? Incidentaly, multipath propagation
        problem in communication engineering is almost the same one
        - low S/N ratio is not a problem but crossinterference is!

        Cheers

        Drago

