[Back]

A. Lorenz, M. Kozek:
"Comparison of Polynomial and Neural Network Models for Information Extraction from a Data Base of Measurements";
Talk: 2008 International Simulation Multi-conference, Edinburgh, Scotland; 2008-06-16 - 2008-06-19; in: "Proceedings of the 2008 Summer Simulation Multiconference", (2008), ISBN: 1-56555-320-9; 8 pages.

This article deals with the optimization and validation of polynomial and Neural Network (NN) models for statistical values extracted from a data base of measurements. These measurements are taken from earthmoving vehicles to characterize the performance with respect to different machine parameters and lead to a large amount of raw data in a high dimensional input space. After the condensation to haracteristical values, the problems of sparse and inaccurate data
have to be dealt with. Simulations on appropriate test data emonstrate the functionality but also the limitations of each method. Experimental results confirm the conclusions of the simulation. Due to the utilized Bayesian regularization techniques, NN can deal with noisemuch better than polynomials. Polynomial models are only useful for simple functions like linear relations.

data mining, modeling, Neural Networks (NN), Bayesian regularization

Project Head Martin Kozek:
Virtuelle Messdatenerfassung (KE 2009 offen)