Title:

Dobór zmiennych do zagregowanych modeli dyskryminacyjnych z wykorzystaniem algorytmów genetycznych

Group publication title:

Prace Naukowe Akademii Ekonomicznej im. Oskara Langego we Wrocławiu

Title in english:

Selection of variables into aggregated discriminant models with usage of genetic algorithms

Creator:

Dudek, Andrzej ; Kurzydłowski, Adam

Description:

Prace Naukowe Akademii Ekonomicznej we Wrocławiu; 2007; nr 1189, s. 147-155

Abstrakt:

In discriminant analysis studies models using single classification trees are often replaced by models aggregating partial models into one multiple model. Breiman [ 1996] showed that significant reduction of classification error can be achieved when independency of partial models is fulfilld. So key role in such approach plays appropriate selection of objects or variables of subsets of training set. Among methods of selection of objects into aggregated models the most often used are: boosting, bagging, adaptive bagging, arcing, windowing. Among methods of selection of variables into aggregated models Correlation-based Feature Selection (CFS) developed by Hall [2000] was most effective. Gatnar [2003] proposed Correlation-based Feature Selection based on Hellwig Heuristic (CFSH) method and empirically showed that CFSH gave smaller classification errors that CFS. In this paper some modification of CFSH is proposed and genetic algorithms are used to find best subsets of variables in partial models. Classification errors for CFFS, CFSH and modified method are compared on datasets from University o f California Repository o f Machine Learning.

Publisher:

Wydawnictwo Akademii Ekonomicznej im. Oskara Langego we Wrocławiu

Place of publication:

Wrocław

Date:

2007

Resource Type:

artykuł

Language:

pol

Relation:

Prace Naukowe Akademii Ekonomicznej im. Oskara Langego we Wrocławiu; 2007; nr 1189 ; Zastosowanie metod ilościowych

Rights:

Wszystkie prawa zastrzeżone (Copyright)

Access Rights:

Dla wszystkich w zakresie dozwolonego użytku

Location:

Uniwersytet Ekonomiczny we Wrocławiu

Coverage:

Projekt dofinansowany ze środków budżetu państwa, przyznanych przez Ministra Nauki w ramach Programu Społeczna odpowiedzialność nauki II. Tytuł projektu: Nauka dla Społeczeństwa: Prace Naukowe AEW w otwartym dostępie (2005-2008)