Diese Einträge stammen aus der Österreichischen Dissertations-Datenbank in 
unredigierter Form

Pagination: 133 p.
Publikationsdatum: 1998
Sprache: deutsch
Affiliation: UK00110; Institut fuer Mathematik;
Universitaet Klagenfurt
Begutachter: Pilz, J.
Viertl, R.
Akad. Grad: Thesis (Dr. rer.nat.)
Klassifikation: G0950 (Statistik); G0920 (Mathematik)
Schlagworte deutsch: neuronale Netze; Geostatistik; lineare Modelle; raeumliche 
Statistik;
Schlagworte englisch: artificial neural networks; spatial statistics; kriging; 
linear models; splines; repression models; software engineering;

Zusammenfassung englisch: This work deals with the question whether or not 
artificial neural networks (ANN) are a valuable choice amongst other classical 
as well as modern techniques in spatial statistics. An introduction to basic 
terms of ANN is followed by a comparison of ANN and statistical methods (linear 
and nonlinear regression models) based on a review of recent literature. The 
next section covers special requirements of spatial prediction and finally 
suggests a combination of ANN and ordinary kriging. Using this approach global 
trend is eliminated by ANN and smoothing of the residuals is done by kriging. 
Another part of this work is dedicated to implementation and application of ANN 
in a freely available statistical software framework called 'R'. Several 
software libraries and their use are described, some of them had to be ported 
to the 'R' environment. Comparison of ANN, kriging, thin plate spline 
regression and local fitting is carried out using a freely available data set 
of US coal parameters. A CD-ROM containing both the text (HTML formatted) as 
well as the data set and a collection of used software is included