Naslov
Synthetic method back-propagation analytic hierarchy process

Autori
Kliček, Božidar ; Dobša, Jasminka ; Hunjak, Tihomir

Vrsta, podvrsta i kategorija rada
Sažeci sa skupova, sažetak, znanstveni

Izvornik
Abstracts of KOI'98 / Scitovski, Rudolf - Rovinj : Društvo za operacijska istraživanja, 1998

Skup
7-th International Conference on Operational Research KOI'98

Mjesto i datum
Rovinj, Hrvatska, 30.09.1998. - 02.10.1998

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
AHP; neural networks; machine learning

Sažetak
The article shows limitations of the method AHP which are the consequences of linear dependance of the output variable upon input variables. For this purpose we compare two ways of modelling profit function : using neural network and using AHP. Each of the quoted methods has some advantages: neural networks have a possibility of learning from data and modelling on linearity ; on the other hand, AHP method parameters are modelled by people - experts. The data base of cases solving the problem of supplying credits in banks is used to prove the limitations of AHP method. It is proved that AHP is the special case of Neural Network Back - Propagation with hierarhic levels where the identity is the transfer function . Further, it is showed that it is possible to reduce the hierarhical AHP network to the neural network with only one hidde level and the identity as the transfer function. The exactness of modelling the classical BP neural network is compared with the linear AHP network. Furthermore, it is showed that for the typical case of supplying credits AHP network is less exact than BP network. The inexactness of BP network is the result of noise i learning data and the imprecision of modelling the profit function ; on the other hand, the in exactness of AHP is the result of noise in data that experts give and on linear dependance of the profit functio upon data. To unite good properties of AHP a d BP method the example is given in which the Neural Network

Izvorni jezik
Engleski

Znanstvena područja
Informacijske i komunikacijske znanosti

POVEZANOST RADA